Two-phase shower immersion cooling system for data centers: Startup sequence

Abstract

The present invention provides a cooling system and a pre-cooling method that can improve efficiency. [Solution] A cooling system and a pre-cooling method are disclosed. The cooling system includes a refrigerant chamber and a condenser. The condenser includes a condensing channel. The cooling system is configured to perform a startup sequence before the cooling operation. The startup sequence is configured to remove at least air from the refrigerant chamber.

Description

[Technical Field]

[0001] 1. Field of Invention

[0002] This application relates to a cooling system and a pre-cooling method, and more specifically, to a cooling system and a pre-cooling method that can improve efficiency. [Background technology]

[0003] 2. Description of Conventional Technology

[0004] Unless otherwise specified herein, the methods described in this section are not prior art to the claims of this application, nor are they recognized as prior art by their inclusion in this section.

[0005] Existing embodiments of two-phase immersion cooling (2P-LIC) technology have several structural and operational deficiencies that limit their practicality in high-density hyperscale data center environments. As shown in Figure 25, the primary structural limitation stems from the method used for maintenance access. Because printed circuit boards (PCBs) must be lifted vertically from the coolant immersion tank for service, these systems inherently have very low vertical packing density ("1"). This vertical removal mechanism wastes valuable rack space because it requires large clearance for PCB movement.

[0006] Furthermore, this vertical service model complicates the mechanical design, often requiring cumbersome, electrically operated top covers to access components below the liquid level. Such heavy covers must also seal the steam space, potentially generating significant steam pressure during normal operation, adding unnecessary complexity, cost, and maintenance burdens to the chassis design, as well as reducing the overall efficiency of the system's service.

[0007] The inherent complexity and environmental risks of the maintenance process pose significant operational challenges ("2"). When PCBs are lifted for service, they become saturated with liquid coolant, making dripping and spilling refrigerant unavoidable during this process. In addition to the mechanical need to lift the components, the volatility of the working fluid increases vapor pressure when operating GPUs / CPUs mounted on the PCBs, necessitating the installation of robust covers ("4") over the coolant tanks to prevent the release of uncontrolled leaked gases, especially when the coolant contains perfluoroalkyl and polyfluoroalkyl compounds (PFAS). This is a critical challenge for the industry, as leaks of these regulated substances raise serious environmental, health, and rapidly evolving regulatory compliance (PFAS compliance), thereby increasing operational overhead and responsibility.

[0008] Conventional 2P-LIC containment methods can also lead to reduced resource utilization. In particular, considering the regulatory requirements of PFAS compliance, which necessitate a heavy lid to seal the pressurized refrigerant vapor, the chassis typically needs to be filled to the refrigerant tank ("3") to completely immerse the components in order to ensure sufficient buffer for vapor expansion and pressure stability within the system. Furthermore, a large volume of internal chassis space must be reserved exclusively for vapor circulation in order to manage and prevent pressure runaway from the evaporating fluid, thus reducing the effective volume available for computing hardware.

[0009] This large enclosed steam volume directly impacts the design of the condenser elements. Systems that rely on passive or diffusive condensation mechanisms (i.e., as noted in “5”, where the refrigerant vapor permeates the lateral side of the condenser rather than passing through it) inherently require condensers with very large surface areas. Consequently, the support infrastructure also needs to accommodate large spaces for steam circulation, reducing the overall horizontal space utilization rate (“6”) within the chassis and rack footprint.

[0010] Finally, the need for vertical alignment when lifting the PCB impacts wiring and connectivity. Because the rear of the chassis is often blocked to support vertical service paths ("7"), all network and power cables must be routed to the bottom. This structural compromise necessitates the use of long cables to ensure vertical lifting clearance. Long network cables may prevent the use of short, high-speed fiber optic connections, potentially complicating or limiting cable management requirements unnecessarily. Similarly, long power cables introduce undesirable resistance losses, thereby reducing overall power efficiency.

[0011] In addition to these implementation challenges, the rapid increase in power density of modern computing components is exposing fundamental heat dissipation bottlenecks in existing cooling technologies. Power consumption of GPUs (or other derivative devices such as NPUs, TPUs, etc.) is projected to increase fourfold from 1.2kW per chip to 3kW-5kW in the next 4-7 years, meaning current cooling solutions will soon exceed their thermal limits. For example, in state-of-the-art cooling technologies such as D2C (Direct-to-Chip) liquid cooling, the overall thermal resistance R from the transistor junction to the surface in contact with the coolant is a significant bottleneck. H While the cooling efficiency is 0.012-0.013°C / W / chip, at expected power levels of 3kW-5kW per chip, the temperature difference (ΔT) between the junction and the coolant can reach up to 40°C or 60°C (0.013 × 3000 = 39, 0.012 × 5000 = 60). Such a significant temperature increase makes it difficult to achieve the target chip performance while maintaining desirable reliability. This indicates that advances in cooling technology alone will not be sufficient to cope with the rapid increase in power density in the future, and that an innovative approach will be necessary.

[0012] Whether air-cooled or liquid-cooled, a fundamental limitation of current thermal solutions is their reliance on integrated heat spreader (IHS) designs. In IHS designs, the transistor junction T rA long series thermal resistance R from the substrate, through multiple thermal conduction materials (TIM1 and TIM2), the IHS itself, and finally to the heat sink until reaching the final cooling medium H occurs. An effective IHS greatly depends on R H_Heatsink→Coolant being significantly higher than the R of all other subsections H (since the heat sink can function properly). This assumption is easily met when the final cooling medium is air. However, currently, liquid coolants are becoming popular in AI server racks, and even at a power density of 1.5 kW / AI chip, R H_Coldplate→Coolant is no longer a clearly dominant factor. To cope with a significant future increase in power per chip, it is necessary to significantly reduce the overall R H , and the R H subsection that should be noted first is the final stage R H_Coldplate→Coolant which currently has the highest value. From this perspective, single-phase cooling technology, even in its liquid-cooled version, is already approaching its limits in current-generation (2025) AI servers, leaving little room for improvement. The future direction for significantly reducing R H_Coldplate→Coolant is undoubtedly two-phase cooling technology. This is because in two-phase cooling, the latent heat of vaporization at the boiling point of the coolant is utilized, resulting in R H_Coldplate→Coolant@BP→0 (R H_Coldplate→Coolant@BP approaches 0 at the boiling point of the coolant), which is optimal for coping with a rapid increase in power consumption per chip.

[0013] However, R H_Coldplate→Coolant@BP→0 means that the basis of IHS operation (R H→Coolant is the main R H component) is lost, and in a two-phase cooling environment for high-power GPU / CPU / NPU, adding the IHS and heat sink itself may ultimately have the opposite effect. This is due to two important factors. First, the thermal conductivity G H of materials such as copper (Cu) and aluminum (Al) is only 1.5 to 2.5 times that of silicon (Si). Second, in a two-phase immersion scenario, due to the latent heat absorption during evaporation (liquid phase → gas phase change), the thermal resistance from the liquid to the coolant at the boiling point (RH→Coolant ) approaches zero. As a result, the additional thermal resistance provided by the IHS, TIM, and heatsink offers no benefit, and the total thermal resistance R H,Total This unnecessarily increases the risk of thermal management problems.

[0014] Therefore, it is necessary to improve the conventional technology. [Overview of the Initiative]

[0015] Therefore, the main objective of this application is to provide a cooling system and a pre-cooling method that improve upon the shortcomings of the prior art.

[0016] One embodiment of the present invention provides a cooling system comprising a refrigerant chamber and a condenser, the condenser comprising a condensation channel, the cooling system being configured to perform a startup sequence before the cooling operation, the startup sequence being configured to remove at least air from the refrigerant chamber.

[0017] One embodiment of the present invention provides a pre-cooling method applicable to a cooling system, the pre-cooling method comprising the step of performing a start sequence before the cooling operation of the cooling system, the start sequence being configured to remove air from at least one refrigerant chamber, the cooling system comprising the refrigerant chamber.

[0018] One embodiment of the present invention provides a cooling system comprising a refrigerant chamber into which a liquid refrigerant is injected, and a recirculation and condensation unit comprising a condenser and a refrigerant reservoir, the recirculation and condensation unit being located outside the refrigerant chamber and configured to recirculate the liquid refrigerant, thereby establishing circulation of the liquid refrigerant between the refrigerant chamber and the recirculation and condensation unit.

[0019] One embodiment of the present invention provides a cooling system comprising a refrigerant chamber and a condenser, the condenser comprising a condensing channel, wherein refrigerant vapor flows through the condensing channel of the condenser and condenses into a liquid refrigerant on the inner surface of the condensing channel.

[0020] These and other objectives of the present invention will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments shown in various figures and drawings. [Brief explanation of the drawing]

[0021] [Figure 1] This is a schematic diagram of a cooling system according to one embodiment of the present invention. [Figure 2] This is a schematic diagram of a server rack according to an embodiment of the present invention. [Figure 3] This is a schematic diagram of a cooling system according to one embodiment of the present invention. [Figure 4] This is a schematic diagram of a slanted fin according to one embodiment of the present invention. [Figure 5] This is a schematic diagram of a slanted fin according to one embodiment of the present invention. [Figure 6] This is a schematic diagram of a cooling system according to one embodiment of the present invention. [Figure 7] This is a schematic diagram of an immersion pocket according to one embodiment of the present invention. [Figure 8] This is a schematic diagram of the heating surface and steam bubble of the present invention. [Figure 9] This is a schematic diagram of an immersion pocket according to one embodiment of the present invention. [Figure 10] This is a schematic diagram of a rotating element according to one embodiment of the present invention. [Figure 11] This is a schematic diagram of the convection wall in an immersion pocket according to one embodiment of the present invention. [Figure 12] This is a schematic diagram of the convection wall in an immersion pocket according to one embodiment of the present invention. [Figure 13] This is a schematic diagram of the heated surface or microstructure within an immersion pocket according to one embodiment of the present invention. [Figure 14] This is a schematic diagram of the heated surface or microstructure within an immersion pocket according to one embodiment of the present invention. [Figure 15] This is a schematic diagram of the heated surface or microstructure within an immersion pocket according to one embodiment of the present invention. [Figure 16] This is a schematic diagram of an immersion pocket formed by inclined fins according to one embodiment of the present invention. [Figure 17] This is a schematic diagram of multiple blowers according to one embodiment of the present invention. [Figure 18] This is a schematic diagram of a blowerless cooling system according to one embodiment of the present invention. [Figure 19] This is a schematic diagram of the temperature profile in the Z direction. [Figure 20] This is a schematic diagram of the post-cooling process (stop sequence) PDS1 according to one embodiment of the present invention. [Figure 21] This is a schematic diagram from the perspective of an IT engineer showing how to remove a single PCB by sliding it. [Figure 22] This is a schematic diagram of a cooling system according to one embodiment of the present invention. [Figure 23] This is a schematic diagram of the post-cooling process (stop sequence) PDS2 according to one embodiment of the present invention. [Figure 24] This is a schematic diagram of the boot sequence PUS according to one embodiment of the present invention. [Figure 25] This is a schematic diagram of a conventional two-phase liquid immersion cooling (2P-LIC) system. [Figure 26] This is a schematic diagram of a conventional condenser system. [Figure 27] This is a schematic diagram of the heated surface or microstructure within an immersion pocket according to one embodiment of the present invention. [Modes for carrying out the invention]

[0022] The technical features described in the embodiments of the present invention can be mixed or combined in various ways, as long as they do not contradict each other.

[0023] Two-phase immersion cooling (2P-LIC) systems primarily achieve their cooling effect by converting the liquid coolant into vapor at the surface in contact with the high-temperature GPU / CPU. Once the coolant turns into vapor, its heat resistance becomes extremely high. If left unchecked, as in conventional 2P-LIC embodiments, this vapor can form bubbles and cover a large portion of the GPU / CPU surface. Surfaces covered by this vapor (VCS) lose their exposure to the liquid coolant, resulting in insufficient cooling. Even worse, the proportion of vapor-covered surface increases with increasing localized power consumption. For example, assuming the GPU chip surface is divided into 500x500 sections, sections with high power consumption will have a higher VCS ratio, less exposure to the liquid coolant, insufficient 2P cooling, higher surface temperatures in those sections, and a further increase in the VCS ratio—creating a vicious cycle.

[0024] Furthermore, when a PCB (printed circuit board) is inserted vertically into a refrigerant tank, the pressure from the liquid refrigerant increases with immersion depth. This increase in liquid refrigerant pressure also raises the boiling point of the refrigerant (vapor pressure is defined as being equal to the local pressure), potentially causing the operating temperature of the GPU / CPU to rise with immersion depth. In the field of AI computing, the speed and efficiency of data access / exchange are paramount. PCBs are becoming larger to minimize the time and power required for data traffic (to accommodate more data traffic locally within a single PCB rather than sending it from the PCB to the backplane or outside the rack). This means that the immersion depth increases, and the rise in boiling point becomes greater.

[0025] All of these effects are highly undesirable, and it would be beneficial if the 2P-LIC system incorporated inherent / organic mechanisms to address the aforementioned problems, such as vapor purging / washing / bubble decomposition and liquid refrigerant circulation.

[0026] One way to counteract the effect of "boiling point rise due to immersion depth" is to reduce the effective density of the liquid refrigerant-refrigerant vapor mixture by increasing the vapor-to-liquid ratio of the refrigerant, and to minimize the adverse effect on the cooling capacity of the 2P-LIC by using ultrasonic means to break down large vapor bubbles into very small bubbles. An alternative to counteracting the effect of "boiling point rise due to immersion depth" is to divide a single large "general-purpose refrigerant immersion tank" containing many PCBs into many small, vertically partitioned "local immersion tanks," each covering one or a few chips, thereby dividing the overall "immersion depth" into multiple short "virtual immersion depth" segments. This suppresses the physical "depth" and the associated pressure rise due to gravity.

[0027] To maximize contact between the refrigerant and the heating surface, a microsurface structure can be formed directly on the back surface of the silicon substrate (opposite the FET circuit side), creating a local surface tension profile (within the liquid refrigerant) and a local adhesion profile (between the liquid refrigerant and the surface of the microstructure), allowing the liquid refrigerant and refrigerant vapor to form a micro liquid refrigerant supply driver (continuously drawing in new liquid refrigerant) and pipe (maintaining an orderly flow of liquid refrigerant and minimizing turbulence in the refrigerant flow), a micro vapor outlet (moving vapor in the opposite direction to the incoming refrigerant, away from the heating surface, and not competing with the incoming refrigerant supply), and a micro vapor conduit (moving parallel to the heating surface, from the refrigerant supply pipe to the vapor outlet pipe).

[0028] Figure 1 is a schematic diagram of a cooling system 10 according to one embodiment of the present invention. The cooling system 10 can be applied to server systems with high / massive computing power (e.g., deployed in a data center). Specifically, the cooling system 10 can be applied to the chassis within the server rack of a server system. The cooling system 10 includes a refrigerant chamber CSC, a plurality of refrigerant injectors 112, a plurality of air-vapor return vents 114, a liquid refrigerant collector 110, and a recirculation and condensation unit RCU.

[0029] In the present invention, the following terms may be used interchangeably: the relationship between the refrigerant chamber and CSC, the evaporation space (i.e., the space in which the refrigerant evaporates) and VPS, the condensation space (i.e., the space inside the condenser) and CDS, and the post-condenser space (i.e., the space immediately after the condenser, e.g., the space above the refrigerant tank) and ACS.

[0030] In Figure 1 and some of the figures in this application, X (pointing to the right) and Y (pointing towards the back or out of the paper) represent the horizontal direction, and Z (pointing downwards) represents the vertical direction or the direction of gravity. Here, G represents gravity. Downward may mean towards the Earth, which is the direction of gravity, or it may mean towards the sky, which is the opposite direction of gravity.

[0031] Figure 2 shows three different configurations of the refrigerant chamber (CSC) and recirculation and condensation unit (RCU) located within a server rack. The refrigerant chamber (CSC) can be located within the server rack of the server system or within a chassis housed on top of the server rack.

[0032] Returning to Figure 1, multiple circuit boards 120 are arranged vertically within the refrigerant chamber CSC. In this application, "circuit boards 120 are arranged vertically" or "circuit boards 120 are facing each other vertically" includes cases where the circuit boards 120 are tilted by a predetermined angle, and that angle may be within a specific (narrow) range, for example, within the range of [-15°, +15°]. Multiple heat-generating components 122 are arranged on the circuit boards 120.

[0033] The heat-generating component 122 generates heat during operation. The heat-generating component 122 may be a CPU, GPU, NPU, TPU, or XPU, where C / G / N / T stands for Central / Graphics / Neural-network / Tensor, and PC stands for Processing Unit. XPU may stand for Extension Processing Unit or other types of Processing Unit. Component 112 may also refer to DRAM, PMIC, network interface IC, etc.

[0034] The refrigerant injector 112 is located at the top of the refrigerant chamber CSC. The refrigerant injector 112 is configured to inject / shower liquid refrigerant into the refrigerant chamber. In one embodiment, the refrigerant injector 112 is a showerhead, includes a showerhead, or functions like a showerhead, generating multiple liquid refrigerant vapors in multiple flow directions. Meanwhile, the liquid refrigerant collector 110 is located at the bottom of the refrigerant chamber CSC. In addition to horizontally oriented plates, the liquid refrigerant collector 110 may further include a network of trenches and / or grooves that help collect / accumulate liquid refrigerant that has not evaporated after passing through the heat-generating components 122 and the circuit board 120.

[0035] In the embodiment shown in Figure 1, the air-vapor return vent 114 is located at the top of the refrigerant chamber CSC, but is not limited thereto. The air-vapor return vent 114 is connected to a recirculation and condensation unit RCU and is configured to deliver a mixture of air and refrigerant vapor from the refrigerant chamber CSC to the recirculation and condensation unit RCU. In practice, the air-vapor return vent 114 can be placed in an appropriate location within the refrigerant chamber CSC such that refrigerant vapor is collected before (or preferentially before) air and efficiently transported to the RCU. For example, since vapor is heavier than air, the air-vapor return vent 114 can be placed at the bottom of the refrigerant chamber CSC.

[0036] The liquid refrigerant can be injected / showered into the refrigerant chamber CSC, for example, via a refrigerant injector 112. The liquid refrigerant flows downward (from top to bottom) due to gravity, passing over the circuit board 120 (specifically, the heat-generating components 122). As the liquid refrigerant flows towards the bottom of the CSC, some of it may absorb heat from the components 122 on the circuit board (e.g., PCB) 120 and evaporate as refrigerant vapor. Any liquid refrigerant that does not evaporate falls / drops to the bottom of the refrigerant chamber CSC and is collected by the liquid refrigerant collector 110.

[0037] Furthermore, in order to properly cool the last (bottommost) component 122 in the CSC under all conditions, it is essential to supply an excess of refrigerant from the injector 112. This ensures that a sufficient amount of refrigerant flows through every path, regardless of whether the components along the path are generating a large amount of heat or not at that moment. Therefore, under normal operating conditions of the refrigerant collector 110, a certain level of "excess refrigerant" exists. The specific level of this "excess refrigerant" is a factor determined by the system designer.

[0038] The recirculation and condensation unit RCU recirculates the liquid refrigerant collected by the liquid refrigerant collector 110 and recovers refrigerant (by condensation) from the vapor recovered from the refrigerant chamber CSC.

[0039] Unlike existing immersion cooling methods, the circuit board 120 is not physically immersed in the liquid refrigerant. Instead, the cooling system 10 employs a shower cooling method, and the circuit board 120 can be thought of as being showered with liquid refrigerant during the cooling process / operation. In other words, during the operation (or cooling operation) of the cooling system 10, the cooling system 10 includes a fluid liquid refrigerant that flows over or through multiple circuit boards 120 due to gravity. That is, during the (cooling) operation of the cooling system 10, gravity causes the fluid liquid refrigerant to continuously flow from the top to the bottom of the refrigerant chamber CSC. Furthermore, the space between adjacent circuit boards is filled with a mixture of air and refrigerant vapor. In this case, the amount of liquid refrigerant required per chassis is significantly reduced.

[0040] In one embodiment, as shown in Figure 16, a structure including an array of upwardly inclined fins is applied to the heating surface of component 122. In this embodiment, an immersion pocket 125 can be formed between vertically adjacent upwardly inclined fins. A virtual immersion depth D in the X / horizontal direction is established when the refrigerant flow rate is sufficient to maintain filling the immersion pocket 125. There are two major differences between the immersion pocket 125 and the immersion tank in a 2P-LIC system. First, the immersion depth is horizontal and perpendicular to gravity. Therefore, all adverse effects related to gravity (e.g., vapor being pressed against the heating surface of component 122 by G, or the boiling point rising with increasing immersion depth) are virtually eliminated. Second, since the immersion pocket is not permanent but formed temporarily and ad hoc, the immersion pocket 125 disappears when the refrigerant shower is stopped, and all problems associated with the 2P-LIC immersion tank become irrelevant.

[0041] These two fundamental differences mean that the immersion pocket 125 is called "virtual," in contrast to conventional immersion tanks, which are "physical."

[0042] The recirculation and condensation unit (RCU), together with the refrigerant chamber (CSC), establishes liquid refrigerant circulation as shown in Figure 3(a) and refrigerant vapor circulation for the cooling system 10 as shown in Figure 3(b).

[0043] For the circulation of liquid refrigerant, the recirculation and condensation unit (RCU) includes a refrigerant reservoir / tank 102 and pumps 101 and 105. Pump 105 is connected to a liquid refrigerant collector 110 and configured to discharge the collected liquid refrigerant from the liquid refrigerant collector 110 to the refrigerant reservoir / tank 102. A filter 104 is optionally placed between pump 105 and the refrigerant reservoir / tank 102 to filter out dust or other types of impurities / contaminants from the collected liquid refrigerant. Pump 101 is connected between the refrigerant reservoir / tank 102 and a refrigerant (distribution) pipe 111. Pump 101 is configured to force the liquid refrigerant in the refrigerant reservoir / tank 102 through the refrigerant (distribution) pipe 111 toward a refrigerant injector 112, which injects / showers the liquid refrigerant into the refrigerant chamber (CSC).

[0044] Furthermore, the cooling system of the present invention may include a liquid level sensor 115 positioned within the liquid refrigerant collector 110 and configured to monitor the liquid level corresponding to the liquid refrigerant collector 110. If the liquid level in the liquid refrigerant collector 110 (sensed by the liquid level sensor 115) is too low / too high, it indicates that there is too much / too little heat generated by the heat-generating component 122 and too much / too little evaporation of the liquid refrigerant. Therefore, the injection rate (or flow rate) of the liquid refrigerant injected into the refrigerant chamber CSC through the refrigerant injector 112 needs to be increased / decreased to compensate for the increase / decrease in the evaporation rate of the liquid refrigerant. In other words, the amount of liquid refrigerant injected into the refrigerant chamber can be controlled / adjusted according to the liquid level sensing result of the refrigerant collector 110 by the liquid level sensor 115.

[0045] In this application, unless otherwise specified, the terms "refrigerant reservoir" and "refrigerant tank" are used interchangeably.

[0046] For refrigerant vapor circulation (see Figure 3b), the recirculation and condensation unit RCU includes a condenser 108 and a refrigerant reservoir 102. Refrigerant vapor circulation involves vaporizing the liquid refrigerant in the refrigerant chamber CSC, sending the vapor to the recirculation and condensation unit RCU located outside the refrigerant chamber CSC, condensing the refrigerant and storing it in the refrigerant reservoir 102, and then recirculating it back into the refrigerant chamber CSC. In one embodiment, the cooling system 10 may optionally include a blower 107. The blower 107 is configured to generate an airflow from the refrigerant chamber CSC to the refrigerant condenser 108, thereby 1) discharging a mixture of air and refrigerant vapor into the refrigerant condenser 108, 2) establishing a pressure difference between the refrigerant chamber CSC and the refrigerant condenser 108, and 3) reducing the air pressure (P) within the refrigerant chamber CSC. CSC) By lowering the boiling point BP of the refrigerant (this promotes the evaporation of the refrigerant), and also, P CSC When the pressure is slightly lower than atmospheric pressure, it helps seal the chassis.

[0047] The condenser 108 is connected to the refrigerant reservoir / tank 102. The condenser 108 is configured to condense refrigerant vapor discharged from the refrigerant chamber CSC by the blower 107 into liquid refrigerant, and the condensed liquid refrigerant is dripped and / or stored again in the refrigerant reservoir 102. Furthermore, the condenser 108 includes a condensation channel 108' between the refrigerant chamber CSC and the refrigerant reservoir / tank 102. The airflow from the refrigerant chamber CSC (including a mixture of air and refrigerant vapor) can flow through the condensation channel in the condenser 108. The refrigerant vapor condenses into liquid refrigerant on the inner surface of the condensation channel 108', and the condensed liquid refrigerant can drip / drop into the refrigerant reservoir / tank 102.

[0048] The condensing channel 108' itself may be or include a sealed subspace or space, excluding the inlet / outlet ports to which it is connected. In the embodiment shown in Figure 3b, the condensing channel 108' may, but is not limited to, have the shape of a meandering tube. In some embodiments, the condenser 108 may be or include a plate heat exchanger, as described later. The heat exchanger may include a first channel and a second channel. The first fluid / second fluid flows through the first channel / second channel. Within the heat exchanger, the first fluid is isolated from the second fluid, but the heat or thermal energy carried by the first and second fluids may be exchanged / released from each other. In one embodiment, the first fluid is an airflow from the refrigerant chamber CSC (including a mixture of air and refrigerant vapor), and the second fluid is a facility fluid (e.g., source water from a chiller, cooling tower, etc.), the facility fluid in the present invention is configured to facilitate the condensation of refrigerant vapor into liquid refrigerant. In this case, the condensation channel is the first channel in the heat exchanger connected between the refrigerant chamber CSC and the refrigerant reservoir / tank 102.

[0049] In the present invention, liquid refrigerant circulation basically refers to the refrigerant path from when the refrigerant is injected from the refrigerant injector 112 into the CSC, flows through the circuit board 120 and heat-generating components 122, the remaining unevaporated liquid refrigerant is recovered by the liquid refrigerant collector 110, recirculated by the pump 105, stored again in the refrigerant tank / reservoir 102, and returned to the refrigerant injector 112 by the pump 101.

[0050] In contrast to the "liquid" refrigerant circulation, the "vapor" refrigerant circulation basically refers to a refrigerant path in which the refrigerant evaporates into refrigerant vapor in the refrigerant chamber CSC by coming into direct contact with the heat generated by element 122, and this refrigerant vapor is recirculated (transported) through the condenser 108, and as it passes through the condenser, it recirculates and releases heat into the service fluid in the condensation unit RCU, causing it to condense back into liquid refrigerant. The condensed liquid refrigerant is then stored in the refrigerant tank / reservoir 102. Here, the stored liquid refrigerant is returned to the refrigerant injector 112 by the pump 101.

[0051] Unlike conventional 2P-LIC systems as shown in Figure 25, the condenser 2530 may have a radiator-like structure as shown in Figure 26 (cold service fluid CFF enters the end block 2531 through the bottom port 103, and heated service fluid WFF exits the end block 2531 through the top port 104, where heat exchange occurs between the heat pipe 2532 and the fins 2533), with condensation occurring on the outer surface of the fins 2533. In this invention, the cold service fluid CFF enters through port 103, flows outside the condensation channel 108 as shown in 1033, is discharged through port 104, and condensation occurs on the inner surface of the condensation channel (e.g., 108').

[0052] By changing from the conventional 2P-LIC system's passive approach of "waiting for vapor to permeate or diffuse through the condenser" to the present invention's approach of "actively passing refrigerant vapor through the condenser," the maximum heat exchange efficiency is significantly improved, and the space required for the condensation function is drastically reduced. The improved maximum heat exchange efficiency is crucial in addressing rapidly increasing power consumption, and the reduced installation space allows for a compact and highly space-efficient thermal management solution.

[0053] From another perspective, the refrigerant chamber CSC can be considered to include an evaporation space where the refrigerant vaporizes, and the recirculation and condensing unit RCU or condenser 108 can be considered to include a condensation space where the refrigerant vapor condenses into liquid refrigerant. The evaporation space and the condensation space are sealed spaces isolated from each other except for the connecting inlet / outlet ports, and the evaporation space and the condensation space are separated from each other. Furthermore, the recirculation and condensing unit RCU, which includes the condenser 108 and the refrigerant reservoir / tank 102, is located outside the refrigerant chamber CSC, or outside the evaporation space of the refrigerant chamber CSC.

[0054] Furthermore, an airflow (carrying a mixture of air and refrigerant vapor) is generated from the evaporation space to the condensation space. In one embodiment, this airflow may be generated by a blower 107. In another embodiment, the airflow may be generated by the vapor pressure difference between a first high vapor pressure in the evaporation space and a second low vapor pressure in or after the condensation space (details will be described later).

[0055] The cooling system 10 includes a plurality of heat sinks 124 arranged on the heat-generating component 122. Figure 4(a) shows a schematic diagram of a heat sink 124 according to one embodiment of the present invention. Figure 4(b) shows the flow of liquid refrigerant flowing through the heat sink 124 and the flow of refrigerant vapor flowing between the circuit boards 120 according to one embodiment of the present invention.

[0056] The heat sink 124 includes an optional (cooling) plate 126 and a plurality of upward-sloping fins 128 positioned on the cooling plate 126. The plurality of sloping fins 128 are sloped upward toward the ceiling of the refrigerant chamber CSC. Furthermore, holes or gaps are formed inside or between the plurality of upward-sloping fins 128 (details below). When the refrigerant is injected / showered from the top of the refrigerant chamber CSC, the refrigerant first fills the space between the uppermost upward-sloping fin 128 and the cooling plate 126, flows over the tips of the fins, and flows downstream through the holes / gaps formed inside or between the uppermost sloping fin 128. After filling the triangular pocket 125t formed by the uppermost fin 128, the refrigerant fills the trapezoidal space 125 enclosed by the first fin 128, the second fin 128 below the first fin 128, the horizontal line passing through the tips of the second fin 128, and the cooling plate 126, forming a temporary immersion pocket 125. As shown in Figure 4(b), after filling the first immersion pocket 125, the refrigerant continues to flow downstream, passing over the tip of the second fin 128, through the hole / gap, and filling an ad-hoc second immersion pocket 125 directly below the first immersion pocket, etc. Eventually, the refrigerant fills all the immersion pockets 125, and the remaining (liquid) refrigerant falls to the bottom of the refrigerant chamber CSC and is collected by the liquid refrigerant collector 110.

[0057] As shown in Figure 4(a), a plurality of refrigerant immersion pockets 125 are formed, and in the embodiment shown in Figure 4(a), the space between the cooling plate 126 and the upward-inclined fins 128 is filled with refrigerant. The liquid refrigerant in the refrigerant pockets 125 is in direct contact with (part of) the surface of the cooling plate 126 and inclined fins 128 associated with the liquid refrigerant immersion pockets 125. In the embodiment shown in Figure 4(a), a portion of the surface of the cooling plate 126 can be considered a heating surface associated with a heat-generating component. In another embodiment, a portion of the surface of the cooling plate 126 may be removed to allow direct contact between the silicon substrate below and the circulating refrigerant. In any case, in the present invention, a heating surface associated with a heat-generating component generally refers to a surface that can transfer heat from the heat-generating component to the refrigerant.

[0058] In this invention, the heating surface related to the heat-generating component is immersed in a liquid coolant within an immersion pocket.

[0059] Furthermore, while the cooling system 10 is operating, the liquid refrigerant in the liquid refrigerant immersion pocket 125 flows continuously downward or to the bottom.

[0060] The statement that the refrigerant "flows continuously downward" means that the existence of these immersion pockets is temporary and transient, depending on whether they continuously receive refrigerant from above. The refrigerant flow rate also affects the virtual immersion depth D (in the X direction, not the G direction; see Figure 16(a)). A slower refrigerant flow rate reduces the value of D.

[0061] Furthermore, the convection of the liquid refrigerant within the cooling system 10 is significantly enhanced compared to conventional immersion cooling systems. In conventional immersion cooling systems, the liquid refrigerant remains almost static / stationary relative to the circuit board 120 or heat-generating component 122. As is known in the art, good liquid refrigerant convection helps to a) release bubbles of refrigerant vapor from the heated surface and b) dissipate heat generated by the heat-generating component, thereby improving the heat dissipation capacity and heat dissipation efficiency of the cooling system 10 compared to conventional cooling systems.

[0062] As shown in Figure 4(b), during the operation of system 10, streams of liquid refrigerant are injected into the refrigerant chamber CSC via injectors 112. These refrigerant streams flow from top to bottom within the refrigerant chamber CSC. A portion of the liquid refrigerant is accumulated or collected in a continuous layer of immersion pockets 125. The liquid refrigerant in the immersion pockets is heated by heat-generating components and evaporates as refrigerant vapor. Vapor generated by heat-generating components 122 on a single vertically oriented circuit board 120 merges into a single refrigerant vapor stream and passes through the space / channel between adjacent circuit boards.

[0063] Figure 5 shows a front view / cross-sectional view of a plurality of inclined fins 128 (cross-sectional view along the line T-T' shown in Figure 4(b)). In one embodiment, holes 127 can be formed in a plurality of upward inclined fins 128, as shown in Figure 5(a). Liquid refrigerant flows downward through the holes 127 due to gravity G. The holes 127 can be formed by CNC (computer numerical control) drilling. To accommodate the loss of liquid refrigerant by evaporation as it passes through heat-generating components, resulting in a gradual decrease in the amount of refrigerant available to continue the flow downstream, the shape and dimensions of the holes 127 may be tapered from top to bottom to gradually reduce the flow rate of the refrigerant along the direction of gravity G.

[0064] From another perspective, in order to improve the uniform distribution of the refrigerant and direct contact between the refrigerant and the heating surface, the holes 127 and gaps 129 may be arranged alternately across a pair of inclined fins 128.

[0065] In one embodiment, as shown in Figure 5(b), the gaps 129 may be arranged alternately between the inclined fins 128 so that the liquid refrigerant flows smoothly downstream through the gaps 129 and spreads evenly.

[0066] Figure 6 is a schematic diagram of a cooling system 20 according to one embodiment of the present invention. Cooling system 20 is similar to cooling system 10, and thus the same notation is used for the same components. One difference from cooling system 10 is that the condenser 108 is an improved plate heat exchanger (MPHE), or includes an MPHE. Like a normal plate heat exchanger, the MPHE includes two sets of thin channels to accommodate two opposing fluid flow streams. These two sets of channels can be formed between laminates of thin sheets made of a highly thermally conductive material such as stainless steel. These sheets may also incorporate patterns such as double fishbone to enhance heat exchange between the two opposing fluid flows.

[0067] The MPHE structure differs from a conventional plate heat exchanger in that, as shown in Figure 6, the MPHE channel (hereinafter referred to as the first channel) that receives the steam-airflow has openings at both the inlet (the upper opening 116 of the condenser 108) and the outlet (the lower opening 117 of the condenser 108). During RCU operation, two fluid streams flowing in opposite directions enter the condenser 108, and the high-temperature first fluid stream flowing through the first set of channels of the condenser 108 releases heat to the low-temperature second fluid stream flowing through the second set of channels of the condenser 108.

[0068] In system 20, the first high-temperature stream is a mixture of vapor and air extracted from the CSC. This vapor and air stream is transported by blower 107 from the refrigerant chamber CSC to the recirculation and condensation unit RCU via port 107p, then enters a first set of channels in the MPHE 108 via the upper opening 116, flows through the MPHE due to the force of blower 107 and gravity G, and flows toward the lower opening 117. Since port / opening 117 is fully open directly above tank 102, the refrigerant condensate free falls into the refrigerant tank 102.

[0069] When actively blowing a vapor-air mixture from the CSC to the RCU using the blower 107, a return path is required from the RCU back to the CSC. The vapor and air stream can be discharged from the RCU via port 115p and return to the CSC via return pipe 115. Port 115p can be located near the top of the refrigerant tank 102, above the surface of the refrigerant. Note that the desired pressure inside the CSC or VPS may differ from the pressure in the condenser 108 in the refrigerant tank 102 or in the space after ACS. By inserting the regulator 109 into the piping 115, programmable flow resistance can be achieved, and by combining this with the flow control of the blower 107, the desired pressure can be achieved on both the RCU and CSC sides.

[0070] In system 20, the second low-temperature stream is a service fluid stream that is received by and returned by the heat exchanger MPHE or condenser 108. The condenser 108 receives cold service fluid (e.g., raw water) through its lower port 103. As this cold service fluid moves upward along the second set of channels, it absorbs the heat released by the first high-temperature steam-air stream, gradually warming up and returning warm service fluid through its upper port 104.

[0071] The condenser 108, functioning as a multi-plate heat exchanger (MPHE), includes multiple plates (pt). Condensation channels (CC) and facility channels (FC) are formed between the plates. Refrigerant vapor flows downward through the CC channels, and the service fluid flows upward through the FC channels. Latent heat is then transferred from the refrigerant vapor to the service fluid via the plates (pt), causing the vapor to condense.

[0072] When comparing the operation of the condenser 108 in System 20 with that of a conventional 2P-LIC system as shown in Figure 25, it should be noted that both the "contact area between the vapor-air mixture and the service fluid stream" and the "relative flow rate between these two fluid streams" are dramatically increased. This indicates that System 20 can provide a much higher heat exchange capacity (due to a larger contact area) at a much faster heat exchange rate (due to a higher relative flow rate). Furthermore, the space occupied by the heat-exchange type MPHE is significantly smaller compared to the 2P-LIC shown in Figure 25, meaning that the RCU in System 20 requires much less space to perform the same functions as a conventional 2P-LIC system.

[0073] Furthermore, a tube 115 is shown connecting the bottom of the condenser 108 to the bottom of the refrigerant chamber CSS. Through the tube 115, a mixture of residual vapor and air can be fed / returned to the bottom of the refrigerant chamber CSC, which helps to generate an updraft within the CSC and direct the refrigerant vapor generated within the CSC toward the vapor-air return channel 113 below the ceiling of the CSC.

[0074] To maintain the pressure difference between the CSC and RCU, a valve / regulator 109 can be placed inside the tube 115 to control the conduction state (flow resistance or open / closed) associated with the tube 115.

[0075] In Figure 6, VPS represents the "evaporation space" within the refrigerant chamber CSC, and CDS represents the "condensation space" within the condenser 108.

[0076] In addition, cooling systems 10 and 20 utilize inclined fins and the space between the inclined fins and the heating surface associated with the heat-generating component to form an immersion packet, but are not limited to this.

[0077] Figure 7 shows various embodiments of the vertically positioned circuit board of the present invention, with and without immersion pockets. Figure 7(a) shows one embodiment of a vertically positioned circuit board without immersion pockets. The circuit board, including heat-generating components (e.g., CPU / GPU), may simply be positioned vertically within a refrigerant chamber (e.g., CSC). As the refrigerant flows over the heat-generating components, the heat generated by the components is dissipated as long as the heated surfaces of the components are in direct contact with the liquid refrigerant as the refrigerant flows from top to bottom.

[0078] Figures 7(b) to 7(d) show various embodiments of the circuit board including the immersion pocket of the present invention. Figure 7(b) shows an embodiment including an upward-sloping fin, which temporarily forms a virtual immersion pocket as shown / described above.

[0079] Figures 7(c) and 7(d) show immersion pockets 30c / 30d / 30e formed within the casing 32c / 32d / 32e of the present invention. The casing 32c / 32d / 32e includes side walls 34c / 34d / 34e and a bottom 36c / 36d. The casing 32c / 32d / 32e and the circuit board form a pocket (or cavity / chamber) 30c' / 30d' / 30e'. The immersion pocket 30c / 30d / 30e is a pocket 30c' / 30d' / 30e filled with liquid coolant. Heat-generating components (CPU / GPU) are immersed in the liquid coolant within the immersion pocket 30c / 30d / 30e. Additionally, holes or slots 31c / 31d / 31e are formed in the bottom 36c / 36d / 36e. During operation of the cooling system, the liquid refrigerant in the immersion pockets 30c / 30d / 30e flows downstream through the holes 31c / 31d / 31e, and then proceeds, for example, 30e' → 30d' (upper) → 30d' (lower) → ... → to the refrigerant collector 110 located at the bottom of the refrigerant chamber CSC.

[0080] The immersion pocket may encompass all heat-generating components on a single circuit board (as shown in 30c in Figure 7(c)), or it may encompass some or only one heat-generating component on a single circuit board (as shown in 30e and 30d in Figure 7(d), respectively). All of these cases are within the scope of the present invention.

[0081] Furthermore, the heat transfer efficiency of the 2P-LIC system is as shown in Figure 8(a), with respect to the A of the steam-covered surface VCS. V ×T V Product(A V , T V (Minimizing the area and time the heating surface is covered by steam) and the liquid refrigerant A C ×T C Product(A C , T C This can be improved by maximizing the area and time (of which the refrigerant is in direct contact with the heated surface).

[0082] When steam bubbles adhere to and cover the heating surface, direct contact between the refrigerant and the heating surface is blocked, and the thermal resistance R H The thermal resistance (from the heat-generating component to the coolant) increases sharply, and the heat transfer efficiency η (heat transfer efficiency from the heat-generating component to the coolant) decreases significantly. Conventional 2P-LIC systems do not address the problem of the vapor-covered heating surface (VCS), so much space remains uncovered in the process of maximizing the potential of two-phase cooling.

[0083] Before addressing the "heated surface covered in vapor" problem, let's first consider what happened. In both conventional and the present invention, in a 2P-LIC cooling system, when the liquid refrigerant comes into contact with a vertically facing heated surface and evaporates, the vapor is first locked / pressed onto the heated surface by a combination of four forces (atmospheric pressure, the adhesive force of the refrigerant to the heated surface, the surface tension of the refrigerant, and gravity due to the immersion depth). The vapor gathers and forms small bubbles. Before these bubbles separate from the heated surface, direct contact between the refrigerant and the heated surface is blocked within the heated surface surrounded by these bubbles, evaporation stops, and the temperature of the heated surface surrounded by these bubbles locally decreases. BP_CoolantThis exceeds the limit. Simultaneously, at the edges of these vapor bubbles, or slightly outside of them, evaporation continues due to direct contact with the refrigerant, adding vapor to the surrounding bubbles and causing them to grow. Because surface tension is inversely proportional to the bubble diameter, as the bubbles grow larger, the adhesive force begins to exceed the surface tension, causing the liquid refrigerant to penetrate beneath the bubbles. As a result, the bubbles detach (are removed or moved) from the heating surface, and direct contact between the refrigerant and the heating surface is re-established.

[0084] One solution to this steam-induced heat (heat dissipation) problem is to perform steam cleaning and steam purging within the immersion pocket. (Another solution involves using a microsurface structure to form an interleaved microchannel pattern on the heating surface, dedicated to either the incoming refrigerant flow or the outgoing steam flow. This will be discussed later.)

[0085] In this invention, "steam purging" refers to the rapid removal of steam bubbles from the immersion pocket, and "steam cleaning" refers to the rapid and efficient removal of steam bubbles adhering to the surface from the heated surface. On the other hand, "steam cleaning" is the direct contact between the refrigerant and the heated surface, i.e., A C ×T C Maximize the product and reduce the thermal resistance R from the heating surface to the coolant. H By minimizing the heat transfer, the temperature of the heated surface is reliably fixed at the boiling point of the liquid refrigerant. On the other hand, "vapor purging" maximizes the supply of fresh refrigerant by maximizing the supply of liquid refrigerant near the heated surface, thereby increasing the limit of the amount of heat that can be removed by 2P cooling.

[0086] In this invention, the term "vapor purge" is used in two different contexts. One refers to "vapor purging in the immersion pocket with liquid refrigerant," and the other refers to "vapor purging of the entire refrigerant chamber with air" during a power-down sequence (PDS). "Vapor purging of the entire refrigerant chamber with air" will be discussed in detail when discussing the power-down sequence. In the following paragraphs, "vapor purge" refers to "vapor purging with liquid refrigerant" related to the immersion pocket.

[0087] In one embodiment, vapor purging is performed by creating a moderate to high-speed movement of the liquid refrigerant on the heated surface, allowing bubbles generated by the heated surface to be quickly discharged from the immersion pocket.

[0088] One method of vapor purging is self-enhancing refrigerant convection. Self-enhancing refrigerant convection establishes or accelerates the flow of liquid refrigerant by utilizing / enhancing convection, displacing vapor bubbles from the liquid refrigerant within the immersion pocket. Furthermore, the faster movement of liquid refrigerant across the heating surface also creates a scrubbing effect, which in turn facilitates the removal of vapor bubbles adhering to the heating surface.

[0089] In one embodiment, as shown in Figures 4 and 16, a heat sink including upward-sloping fins with alternating holes and gaps can be attached to a heat-generating component. The liquid coolant flows in a zigzag pattern between the fin layers due to gravity, through the alternating path to the next layer of fins below, causing flow on the surface of the heat-generating component. This achieves the objective of "creating a suitable flow of liquid coolant on the heated surface."

[0090] Furthermore, as shown in Figure 16, the lower end of the upward-sloping fins on the sides directs the steam away from the heating surface, further achieving the objective of "quickly discharging the generated bubbles from the immersion pocket."

[0091] In another embodiment, a partition (wall) is provided within the immersion pocket to form first and second liquid refrigerant flows, and one of the liquid refrigerant flows is used to accelerate the second refrigerant flow, thereby more effectively expelling bubbles.

[0092] Figure 9 is a schematic diagram of an immersion pocket 40 according to one embodiment of the present invention. The immersion pocket 40 is formed by / within a casing 42, which includes a vertical wall 44 and a bottom 46 having a refrigerant discharge hole / slot 43. Unlike the immersion pocket of Figure 7, a vertical partition (wall) 48 is provided to divide the volume within the immersion pocket 40 into two subspaces. This partition 48 creates a first refrigerant channel 401 in contact with the vertical wall 44 and a second refrigerant channel 402 in contact with the heating surface within the immersion pocket 40. The refrigerant flows in opposite directions through channels 401 and 402.

[0093] Circulation within the immersion pocket 40 is established by the following two factors: 1) the height L1 of 401 is greater than the height L2 of 402, as shown by ΔL in Figure 9; and 2) the difference in gravity or density of the refrigerant between the refrigerant in channel 401 and the refrigerant in channel 402. Specifically, heat-generating components (e.g., high-performance computing (HPC) GPUs) are immersed in the refrigerant channel 402. Numerous vapor bubbles form and float within channel 402, but no vapor bubbles form within channel 401. As a result, the effective density of channel 402 is significantly lower than that of channel 401, and the gravity FG2 on the channel 402 side of the partition wall 48 is significantly smaller than the gravity FG1 on the channel 401 side. Consequently, a non-zero lateral force is generated along the bottom plane 46 from channel 401 to channel 402, and its magnitude is equal to the gravity difference ΔG = FG1 - FG2. This pushes the mostly transparent (or with a few very small bubbles) liquid refrigerant from channel 401 to channel 402, causing the refrigerant flow in channel 402 to move upward, sweeping out any newly formed vapor bubbles and carrying them to the top of the liquid refrigerant 403, where the bubbles burst and vapor is released.

[0094] In an embodiment like that shown in Figure 9, the casing 42 may be shaped / designed such that there is a non-zero level (channel length) difference ΔL between channels 401 (L1) and 402 (L2). Here, ΔL = L1 - L2 > 0, where L1 / L2 represents the refrigerant overflow level of channels 401 / 402. In the case of the immersion pocket 40, both the density difference and the level / length difference contribute to the gravity difference ΔG, enhancing the circulation of the liquid refrigerant within the immersion pocket 40 and generating a first liquid refrigerant flow CF1 in a first direction (downward) and a second liquid refrigerant flow CF2 in a second direction (upward).

[0095] In one embodiment, the upper part of the first refrigerant channel 401 may be funnel-shaped, which creates a mass / volume difference of refrigerant between the first channel 401 and the second channel 402, contributing to the gravitational difference between them.

[0096] The following discussion regarding Figure 11 will borrow the numbering from Figure 9.

[0097] In embodiments 11(a) to 11(d) of Figure 11, the outer wall 44 is straight rather than funnel-shaped, the end cap (Y direction, not shown directly) is flat and flush with the wall 44, there is no notch 403, the ΔL between channels 401 and 402 is 0, and the partition wall 48 is substantially submerged below the refrigerant surface when the immersion pockets are filled to the rim. Unlike embodiment 40 of Figure 9, during operation (when the immersion pockets are properly filled), channel 401 is connected to channel 402 at both ends of partition wall 48. This allows the refrigerant to flow in a full cycle and convection occurs independently within each pocket.

[0098] In embodiments 11(a) to 11(d) of Figure 11, it should be noted that the strength / size of the liquid refrigerant flows CF1 / CF2 depends on the density difference, and the density difference depends on the volumetric density of the vapor bubbles in channel 402. As the workload of the HPC GPU / CPU increases, the amount of heat generated by the heat-generating components increases, the number of vapor bubbles generated in channel 402 increases, the density decreases further, the magnitude of ΔG increases, and the strength / size of the liquid refrigerant flow (especially CF2) pushing out the vapor bubbles increases. As a result, the entire vapor purging process becomes self-regulating or self-reinforcing, and the vapor discharge strength changes dynamically in real time in response to the heat load in each immersion pocket, thereby independently maintaining the thermal stability of each immersion pocket.

[0099] Furthermore, a rotating element 45 (shown in Figure 10) is provided and positioned between the first refrigerant channel 401 and the second refrigerant channel 402. The rotating element 45 includes a hub 451 and a blade 452 and performs rotational motion. The blade 452 may have a cup-handle shape. When the blade 452 enters the side of the channel 402, its cup-handle shape helps the blade 452 capture / collect rising bubbles. The buoyancy of these captured bubbles generates torque, causing the rotating element 45 to rotate around the hub 451. When the blade 452 enters the side of the channel 402, a downward thrust equal in magnitude but in the opposite direction is applied to the refrigerant flowing through the channel 401. In other words, the rotating element 45 converts previously unused force (upward thrust from vapor bubbles in the channel 402) into a force that accelerates the circulation of the refrigerant (downward thrust of the refrigerant in the channel 401). This further promotes effective self-enhancing convection, improving the effectiveness of both steam purging and steam cleaning.

[0100] To perform these tasks optimally, the tip of the blade 452 typically extends close to both the heating surface (to scoop up most of the rising vapor bubbles in the channel 402) and the inner surface of the wall 44 (to exert a downward thrust throughout the entire column of coolant in the channel 401).

[0101] Figure 11 shows various embodiments of the partitioned immersion pocket of the present invention. In one embodiment, as shown in Figures 11(a) and 11(c), the immersion pocket can contain all or some of the heat-generating components on the circuit board. In another embodiment, as shown in Figures 11(b) and 11(d), multiple immersion pockets are formed on the circuit board, and the number of heat-generating components contained in each immersion pocket is reduced. Figures 11(b) and 11(d) also include a rotating element 45 to promote convection.

[0102] In the embodiment of the partitioned immersion pocket, preferably, the (vertical) projection of the upper refrigerant outlet / slot 411b of the upper immersion pocket is located in the first refrigerant channel (refrigerant channel between the wall and the partition) of the lower immersion pocket. For example, as shown in Figure 11(b), the (vertical) projection of the upper refrigerant outlet 411b of the upper immersion pocket 410b is located in the first refrigerant channel of the lower immersion pocket 420b, and the refrigerant outlet 422b of the immersion pocket 420b is located in the first refrigerant channel of the next lower immersion pocket, and so on.

[0103] Furthermore, an oscillator 47 (shown in Figure 12) may be provided and arranged within the immersion pocket. The oscillator 47 is configured to generate turbulence in the liquid refrigerant, and 1) generates a cleaning action along the heated surface, promoting the removal of vapor bubbles from the heated surface and increasing the direct contact area A between the refrigerant and the heated surface. C ×T C 1) Maximize the efficiency of the liquid refrigerant within channel 402 by breaking down large vapor bubbles into smaller ones.

[0104] In the embodiment shown in Figure 12, one or more transducers 47 are arranged in an immersion pocket, with at least one transducer located near the bottom of the immersion pocket, but this is not limited to this configuration. Furthermore, at least one transducer 47 is located in a refrigerant channel 402, with its primary radiation beam surrounding the heat-generating component.

[0105] In one embodiment, the transducer 47 can vibrate to generate ultrasonic waves, which generate pressure and moving waves of the liquid refrigerant along the surface of the immersed heating component, thereby removing refrigerant vapor bubbles from the heating surface. Furthermore, in one embodiment, the transducer 47 can vibrate to establish standing waves within the refrigerant channel (e.g., 402) or the immersion pocket. In other words, the transducer 47 generates liquid refrigerant waves corresponding to the ultrasonic frequency, and turbulence in the liquid refrigerant can be considered to include liquid refrigerant waves. In yet another embodiment, the transducer 47 can vibrate to continuously generate a number of standing waves with varying durations, such that the nodes and antinodes of different standing waves cancel each other out in the time-averaged overall amplitude response, keeping the overall amplitude response substantially flat across the heating surface.

[0106] Rather than focusing on vapor removal / purging using convection or the generation of liquid refrigerant flow / turbulence, the true objective is to maximize contact between the refrigerant and the heated surface. One approach to achieving this objective is through surface treatment or surface microstructure. By appropriately designing the geometric features and treating the properties of the heated surface, both wettability (direct contact between the heated surface and the liquid refrigerant) and permeability (adequate removal of refrigerant vapor after evaporation) can be improved simultaneously. To achieve the above objective, the surface microstructure needs to simultaneously realize the following three functions: 1) For example, by forming channels of relatively strong or gradually increasing capillary effect, the liquid coolant is smoothly and continuously drawn toward the heat source surface (or the bottom of the microstructure). 2) For example, by forming a region or channel with a very weak capillary effect (compared to 1. above), the refrigerant vapor is released smoothly and continuously from the heat source surface (or the bottom of the microstructure). 3) Form stable lateral steam transfer channels parallel to the heat source surface, and maintain the shape of these steam transfer channels to minimize turbulence in the steam flow.

[0107] The "smooth and continuous" aspect is crucial. Unlike 2P-LICs with a flat heating surface, the need to draw liquid refrigerant towards the heating surface and expel refrigerant vapor from the heating surface results in random, irregular, and turbulent flow. Consider a crowded station during rush hour where many passengers want to board a train and many more want to alight. If the entire train consisted of only one carriage, and all passengers boarded and alighted simultaneously without any order or courtesy, it would be utter chaos, like boiling water in a kettle. However, if the train were divided into several shorter carriages, with people boarding from one end of each carriage and alighting from the other, the boarding and alighting process for all passengers would be quick, smooth, and effortless.

[0108] Therefore, a successful embodiment for achieving the three aforementioned 2P-LIC optimizations (1. maximizing contact between the refrigerant and the heating surface, 2. increasing the rate at which generated vapor is discharged from near the heating surface, and 3. maximizing the supply of refrigerant to the heating surface) can generate three stable, dense, intertwined / linked patterns: the first active area pattern 1: drawing the refrigerant towards the heat source (getting into the car), the second active area pattern 2: releasing / discharging refrigerant vapor from the heat source (getting out of the car), and the third active area pattern 3: the link between the operating areas of the first and second patterns.

[0109] Figure 13 is a schematic diagram of a processed heating surface 521 formed on the back surface of a silicon substrate of a chip (or generally on the back surface of a substrate of a semiconductor device or heat-generating component) according to one embodiment of the present invention. The heating surface 521 is directly seated on the heat conduction layer 520 and therefore comes into direct contact with the coolant. Specifically, Figure 13(a) shows a cross-sectional view of a heat-generating component 522 attached to a PCB, Figure 13(b) shows a detailed cross-sectional enlargement view of a spot 521b on the heating surface 521, and Figure 13(c) shows a detailed top view of the same spot on the heating surface 521. In the present invention, the notation "521" also refers to a microstructure formed on the heating surface of a semiconductor or heat-generating component.

[0110] In one embodiment, the thermal conductive layer 520 may be on the back surface of the silicon substrate of the GPU-CPU die containing HBM, with other dies mounted on the other side (circuit / semiconductor device side). The heating surface 521 can be fabricated directly on the surface of the thermal conductive layer 520 by techniques such as epitaxy, ALD, CVD, epitaxy, and DRIE etching, as shown in Figure 13(b).

[0111] The thermal conductive layer 520 and the refrigerant contact heating surface / layer 521 can be made of a material with high thermal conductivity. In one embodiment, the thermal conductive layer 520 may be a silicon substrate, and the heating surface 521 can be made of SiC (silicon carbide) or AlN (aluminum nitride, 140-320 W / m·K), but is not limited to these. (For reference, the thermal conductivity κ for Si is 150 W / m·K, for AlN it is 140-320 W / m·K, and for SiC it is 120-490 W / m·K)

[0112] The heating surface / heating layer 521 (since 521 is thin, "heating surface" and "heating layer" will be used interchangeably in the following discussion) is composed of a microstructured fabric consisting of multiple protrusions 501. Figures 14(a) and 14(b) show detailed cross-sectional and top views of the heating surface 521 and the protrusions 501. The protrusions 501 can be arranged as a two-dimensional grid, as shown in Figures 13(c), 14(b), and 15(a), or as an alternating array, as shown in Figure 15(b). In the arrangement shown in Figure 14, an inter-protrusion gap 502 and an interstitial space 503 are formed between adjacent protrusions 501. The inter-projection gap 502 of the present invention refers to the (narrow) gap between two adjacent projections and is either in the first direction d1 or the second direction d2 (in the embodiments shown in Figures 14(a) and 14(b), the projections 501 are arranged in a grid of rectangular / square projections with chamfered corners and tapered sides), and is shown as a dashed rounded rectangle in Figure 14(b). The gap space 503 is an open region surrounded (or separated) by four adjacent projections 501 in the array of projections, and is shown as a dashed circle in Figure 14(b). The gap space 503 in Figure 14 (corresponding to the channel 56 in Figure 13(c)) is located at the junction of the four projections 501 and corresponds to the intersection of the volumes defined by two inter-projection grooves, one in the d1 direction and the other in the d2 direction. In other words, the gap space 503 is located in the center of the four surrounding projections 501.

[0113] Note that the notation "502" in Figure 14(b) not only indicates the gap between the protrusions, but may also refer to a (vapor transfer) channel (similar to the path 57 shown in Figure 13(c)). Refrigerant vapor is supplied to the gap space 503 through this channel.

[0114] Preferably, the side walls of the projections (e.g., 501) are tapered gradually inward / rearward from bottom to top, as can be seen in Figures 13(b) and 14(a). This tapered wall means that the gap between the projections 502 narrows downward or towards the base of the projection 501, causing capillary action (adhesive interactions and cohesive forces) to be significantly reinforced towards the base of the projection 501, which helps to draw the liquid coolant from the tip of the projection 501 to its base and to the heat conduction layer 520, which is the heat source.

[0115] Preferably, the projection (e.g., 501) has a geometric shape with convex corners when viewed from above. For example, the projection 501 has an octagonal shape when viewed from above, and the octagonal projection 501 has four short sides facing four gap spaces 503 and four long sides facing four gaps 502 between the projections.

[0116] Furthermore, the first contact area between the projection and the liquid refrigerant within the gap space (e.g., 503) is smaller than the second contact area between the projection and the liquid refrigerant within the gap between the projections (e.g., 502) (e.g., area A is smaller, width W is narrower, or both). In embodiments having octagonal projections (e.g., 501 in Figure 14(b)), the first contact surface is related to the short side of the octagon, and the second contact surface is related to the long side of the octagon. The first / second contact surfaces have opposing contact surfaces across the gap space / gap between the projections, and the distance between the first pair of contact surfaces is longer than the gap between the second pair of contact surfaces.

[0117] Since the capillary force of narrow parallel plates is proportional to / correlated with W / D, the inter-projection capillary force (502s) can be made much stronger than the gap capillary force (503s) by fine-tuning the width W of different contact surfaces and the spacing D between different pairs of contact surfaces. This difference in capillary force extends from the base to the tip of the projection 501, forming two sets of intertwined microchannels. The first set of microchannels (arranged as a first pattern or first array) attracts liquid refrigerant to the heat conduction layer 520 (activation 1) with a strong capillary force, while the second set of microchannels (arranged as a second pattern or second array) releases / exhausts refrigerant vapor from the heat conduction layer 520 with a weaker capillary force (activation 2). The first pattern / array and the second pattern / array are intertwined.

[0118] In short, by optimizing the shape (spacing, area, inclination angle, nanogrooves, etc.) and applying surface treatment (various coatings, surface polishing / roughening, etc.) to fine-tune the forces within these microchannels (capillary action, adhesion, etc.), two different fluids, liquid refrigerant and refrigerant vapor, can achieve their respective purposes (inlet vs. outlet) without interfering with each other. This is almost the same as the aforementioned analogy of getting on and off a train.

[0119] Note regarding the dimensions of these active regions: Considering the miniaturization of semiconductor circuits and the increase in power density today, the pitch of the active regions / dots in the first and second patterns is 1-50 μM, and the height / depth of the protrusions is 5-100 μM, but these are not the only possible values.

[0120] Furthermore, since a heating surface has been obtained that can absorb liquid refrigerant into the heat conduction layer and release refrigerant vapor from the heat conduction layer, a method is needed to link these two operations together.

[0121] As shown by the block arrows in Figure 13(b), heat 51 is generated in the heat conduction layer 520, with the highest temperature at the base of the projection 501, and the temperature decreases along the height direction H as the heat is released to the refrigerant. In Figure 13(b), the refrigerant vapor 52 passes through a gradually narrowing channel defined by the gap between projections 502, with a temperature of TBP_Coolant The liquid refrigerant, preheated to the vicinity, BP_Coolant It is generated when the hotter surface of the projection 501 (indicated by the edge 55) comes into contact with the base at a distance H55 (the periphery of the bottom of the microstructure; "periphery" means approximately below / below H55). Once vapor is generated, due to adhesive forces (between the liquid refrigerant and the surface of the projection 501) and surface tension (within the liquid refrigerant), it collects and forms a pan-shaped vapor tunnel 52 on or around the bottom of the projection 501 within the inter-projection gap 502. The projection surface is T BP_Coolant Where the temperature becomes higher, for example, the edge 55 at height H55 in Figure 13(b), depends on the balance of three elements. These elements are related to the vapor pressure P in the steam tunnel 52. VPR These are the adhesive force or capillary force 53, and the resistance to vapor escaping from the second set of microchannels (56 in Figure 13(c) or 503 in Figure 14(b)), which is mainly caused by the capillary force within the gap space 503.

[0122] Optimizing the microstructure shape produces a series of beneficial thermal effects 1. Specifically, the capillary action within the gap space 503 is weakened, which reduces the vapor pressure (P) within the tunnel 52. VPR The temperature gradient ΔT / H55 (ΔT=T) decreases, and as a result the evaporation edge 55 is lowered. This action reduces the height H55, and as a result the temperature gradient ΔT / H55 (ΔT=T) is lowered across the height H55. 520 -T BP_Coolant ) becomes larger, and ultimately the thermal conductive layer 520(T 520 Liquid refrigerant (T BP_Coolant The heat transfer coefficient to the vicinity (near edge 55) is improved.

[0123] In the ideal case, assuming zero resistance to the vapor discharged from the microchannel 56, the vapor can exist with almost no resistance, and H55 will be very small. This is limited by, firstly, the thermal conductivity of the heating layer 521 and the conductive layer 520, secondly, the friction between the vapor and the walls of the tunnel 52, and thirdly, the system's ability to rapidly supply a sufficient amount of refrigerant at a rate faster than the evaporation rate of the refrigerant.

[0124] In short, one approach to maximizing heat dissipation capacity through increasing gradient is to maximize the gradient ΔT / H55 by minimizing H55. Based on the discussion so far, there are two ways to minimize H55: a) maximizing the capillary force in the gap 502 between protrusions, and b) minimizing the capillary effect in the gap space 503. However, a low capillary effect in the gap space means a large area per 503, while a strong capillary effect in the gap between protrusions means that the protrusions 501 are small and densely packed. These two requirements are contradictory, and the optimal point can be found by trading off one factor for another. In other words, optimizing the design of the microsurface on the heating layer 521 requires detailed modeling and simulation.

[0125] As mentioned above, in addition to minimizing H55, there is another perspective (angle) for maximizing the heat dissipation capacity of the heating surface 521. In a 2P-LIC system, due to the sharp rise in the TS graph (temperature-entropy diagram) at the boiling point, evaporation of the refrigerant actually occurs when the surface temperature of the projection 501 reaches T BP_Coolant Note that this applies only to the portion along edge 55, which is equal to 501. In other words, regardless of the number of protrusions or the size of each protrusion 501, all of their surfaces only serve an auxiliary function, and only edge 55 actually performs the "evaporation" function within the entire array of protrusions. Therefore, the important indicator to optimize is the "evaporation edge length density," i.e., "thermal conduction layer 1 mm 2 This will be the total length (mm) of the edge 55.

[0126] In the case of the planar projection 501, the above criterion refers to "concentrating projections with smaller cross-sectional areas," which is the same direction as maximizing the capillary effect in the gap between projections. In other words, in the case of the flat surface projection 501, "concentrating projections with smaller cross-sectional areas" may be a very favorable direction in exploring the optimal configuration.

[0127] Instead of a flat surface, the capillary force can be further enhanced by etching nano-sawtooth grooves / structures with a pitch of 10-100 nm from the tip to the base of the projection 501, as shown in the top view 14(c). Alternatively, the bottom can be optimized for lateral vapor transfer rather than evaporation by leaving the lower 5-25% of the projection exposed without forming grooves, as shown in the top view d-d' of Figure 14(d), the top view e-e' of Figure 14(e), and the cross-sectional view f-f' of Figure 14(f).

[0128] (As shown in Figures 14(c) and 4(d),) by significantly increasing the capillary force using nano-sawtooth grooves or sawtooth structures, the ability to draw liquid refrigerant from the tip to the base of the protrusion is greatly improved, and the wettability of the protrusion 501 is greatly improved. Note that in the corner of Figure 14(c), instead of nano-grooves, the surface is rounded, which suppresses the capillary effect in the gap space and further distinguishes the first set of microchannels from the second set of microchannels. Furthermore, the nano-grooves are located in the "thermal conduction layer 1 mm" as described above. 2 This will definitely increase the total length (mm) of the edge around 55.

[0129] As described above, when in a stable / balanced state, the surface of the projection 501 around the edge 55 between the vapor-occupied space and the liquid refrigerant becomes close to the boiling point of the refrigerant, the lower surface of the edge 55 (towards the base / including the base) becomes hotter than the boiling point of the refrigerant, and the upper surface of the edge 55 (towards the tip / including the tip) becomes colder than the boiling point of the refrigerant.

[0130] Furthermore, unlike a flat, featureless heating surface, the vapor does not need to dig new paths each time it needs to escape the constraints of adhesion and surface tension. Instead, different paths (57 for lateral vapor flow and 56 for orthogonal vapor flow) are clearly pre-configured within the microstructure and surface features of 521. This holistic design approach results in significant improvements by avoiding or minimizing unnecessary interaction between the incoming liquid refrigerant and the outgoing refrigerant vapor. This effect increases the flow rates of both fluids (liquid refrigerant and refrigerant vapor), and significantly improves the heat treatment capacity per unit area.

[0131] Furthermore, due to the cohesive force or surface tension 54 of the liquid refrigerant, the refrigerant vapor 52 moves laterally along the path / direction 57 toward the region 56 shown in Figure 13(c) or toward the gap space 503 shown in Figure 14(b). Since the distance D2 (which is the diameter of the gap space 503) is greater than the distance D of the inter-projection gap 502, and the first contact surface in the gap space 503 is significantly smaller / narrower than the second contact surface in the inter-projection gap 502, the adhesion force in the gap space 503 is greatly weakened, and the refrigerant vapor bubbles drift out of the gap space.

[0132] Figure 15 shows alternative embodiments of the heating surfaces (top view) 521a and 521b of the present invention. In Figure 15(a), the heating surface 521a is the same as the heating surface 521. The protrusions 501a form a rectangular / square array. The gaps 512 and gap spaces 513 between the protrusions are located within the array of protrusions. Unlike the heating surface 521, the protrusions 501a are circular and convex in top view.

[0133] In Figure 15(b), unlike the array shown in Figure 15(a), the protrusions 501b in Figure 15(b) form an oblique staggered array. Similarly, the gaps 522 and gap spaces 523 between the protrusions are also arranged within the oblique staggered array, which is also within the scope of the present invention.

[0134] By optimizing the heating surface using the method introduced in this invention, H55 can be made very small. Because the thermal conductivity of the heating layer / heating surface 521 is limited, the image of the edge 55 is transmitted through the heating layer 521 and projected onto the underlying heat conduction layer 520. Figures 27(b) and 27(c) show the top surface temperature profiles along the cross sections B-B' and C-C' of Figure 13(b). These profiles are located approximately 1 / 5 of the projection pitch below the plane of the projection base and the 520-521 interface, respectively. Darker colors indicate higher temperatures. These temperature profiles highlight one of the weaknesses of the configuration shown in Figures 13 to 15. That is, because the thermal conductivity is limited, the heating surface 521, which is made of a single material, needs to be thicker in order to uniformly diffuse heat across the interface with the heat conduction layer 520. Furthermore, the unevenness increases as the pitch of the projections increases. In other words, if the density of the protrusions is constant, increasing the ratio of the thickness of the base of 521 to the pitch of the protrusions 501 will improve the unevenness, and conversely, decreasing it will reduce the unevenness. However, if the base 521 becomes thicker, R H Because this increases the cost, "thickening the base 521" is at best not the optimal solution.

[0135] In the cross-sectional view shown in Figure 27(a), an additional high-κ layer 525 is shown inserted between the heat conduction layer 520 and the heating surface 521, and the layer 525 may be made of a material or structure having a much higher conductivity (κ) than the material of the heating surface 521. In one embodiment, the layer 525 is a micro-vapor chamber in which the coolant operates at a boiling point 1 to 5°C higher than the coolant circulating outside the projection 501, dispersing the non-uniformly distributed heat from the heating surface 521 into uniformly distributed heat in the heat conduction layer 520. In another embodiment, the layer 525 is a layer of synthetic diamond with a κ about 7 to 15 times higher than, for example, AlN or SiC.

[0136] The main drawback of adding layer 525 is the cost, and other concerns include material compatibility and manufacturing difficulties. While adding layer 525 is neither inexpensive nor easy, incorporating a high-dielectric constant (High-κ) layer 525 to equalize the heat distribution from the heating surface to the heat conduction layer can significantly improve heat dissipation capabilities, thus occupying a legitimate place in the evolutionary path of the technology based on this patent.

[0137] In Figure 16, vapor purging with a liquid refrigerant can be achieved by the bottom surface of the upward-sloping heat sink fins and by the alternatingly arranged gaps and holes formed in the heat sink including the upward-sloping fins. Figure 16(a) shows a cross-sectional view, and Figure 16(b) shows a front view of the heat sink 224 including upward-sloping fins 201 and 202 having an inclination angle θ. The inclined fins 201 and 202 may be made of metal or plastic (e.g., aluminum, nylon). The inclined fins 201 and 202 can hold the refrigerant in the immersion pocket by covering both ends (in the Y direction, not shown in Figure 16) with caps. Holes 204 may be formed in the inclined fins 201 and 202, and the holes 204 of the inclined fins 201 and 202 are arranged in an alternating pattern / configuration. The alternating pattern / configuration of holes generally means that the vertical projection of the holes 204 of the inclined fin 201 does not overlap with the vertical projection of the holes 204 of the inclined fin 202.

[0138] As the liquid refrigerant fills the pockets between the fins, the vapor bubbles are pushed away from the substrate by the bottom surface of the upper inclined fins. The vapor is scraped off (i.e., removed / purged) immediately after the vapor bubbles are generated, as indicated by pointer 125, rather than drifting across the heated surface and returning to the top of the immersion pockets. This alternating arrangement pattern allows the refrigerant flow to be formed horizontally (Y direction) and vertically (Z direction) 217 ​​as refrigerant flow 206. This facilitates the flow of the liquid refrigerant, which is swept across the vapor generation surface, thus facilitating vapor purging within the immersion pockets.

[0139] Furthermore, the flow of the refrigerant can be enhanced by adding a gap 203 formed between the inclined fins and the heating surface associated with the circuit board, and at the bottom of the immersion pocket. That is, the liquid refrigerant flows downstream through the gap 203 across the heating surface, and the downward flow of liquid refrigerant can wash away the vapor bubbles formed on the heating surface.

[0140] In the embodiment shown in Figure 16, the heating surface is the surface of a silicon substrate (which has a similar function to the thermal conductive layer 520 and may also include a contact layer 521) and is in direct contact with the liquid refrigerant.

[0141] Returning to Figure 6, the key to enabling the cooling operation of the cooling system (e.g., 20) is the airflow (a mixture of vapor and air) generated from the refrigerant chamber CSC to the refrigerant tank passing through the condenser 108. In the embodiment shown in Figure 6, the airflow is generated and can be controlled by the blower 107. The generated airflow is adjustable to adjust the air pressure in the condenser 108 or the dew point of the refrigerant vapor in the condenser 108. Specifically, by controlling the airflow with the blower 107, the air pressure in the condenser 108 or the dew point of the refrigerant vapor in the condenser 108 can be adjusted. Adjusting the air pressure in the condenser 108 to be higher / lower will also raise / lower the dew point of the refrigerant vapor in the condenser 108. Also, since a higher dew point promotes condensation in the condenser 108, the blower 107 can be controlled to raise the pressure in the condenser 108 above atmospheric pressure (e.g., 1.1 ATM).

[0142] Furthermore, the cooling system 20 may include a regulator 109 connected between the condenser 108 and the refrigerant chamber CSC. The blower 107 and the regulator 109 work together to determine the air pressure in the condenser 108, and thus the dew point of the refrigerant vapor can be determined.

[0143] In one embodiment, the vapor partial pressure P is located at the inlet of the condenser 108. VPRWhen the pressure is at its highest, the dew point can be raised by 2 to 6 degrees Celsius by using the blower 107 to increase the air vapor pressure before and after the condenser 108 by 6 to 22 kPa (kilopascals).

[0144] In one embodiment, the vapor partial pressure P is located at the outlet of the condenser 108. VPR When the value is at its lowest, the blower 107 and regulator 109 are T out <=T DP It can be controlled so that -k occurs. Here, k = 3 to 10°C, and T out and T DP This represents the temperature of the airflow at the condenser output (e.g., the temperature between condenser 108 and regulator 109) and the dew point temperature of the vapor at 1 ATM.

[0145] Furthermore, raising the dew point of the refrigerant vapor improves the ease of condensation in the condenser 108, potentially allowing the use of warmer raw water (e.g., 40°C) without the need for a chiller. In other words, by properly designing or adjusting the blower 107 (and regulator 109), it may be possible to save on chiller (or CDU (refrigerant distribution unit)) costs / expenses.

[0146] In other words, the blower 107 and regulator 109 are controlled to adjust the air pressure in the condenser 108 or the dew point of the refrigerant vapor in the condenser, depending on the temperature of the service fluid that the condenser 108 receives.

[0147] While we have so far introduced cooling systems including a single blower, the present invention is not limited to this, and cooling systems including multiple blowers may also be included within the scope of this invention.

[0148] Figure 17 shows a triple-redundant multi-blower system as one embodiment of the present invention. The cooling system of the present invention may include a plurality of blowers, for example, 107a, 107b, and 107c. Valve 107v is provided and positioned at the inlet of the blower connected to the refrigerant chamber CSC, or at the outlet of the blower connected to the condenser 108.

[0149] In one embodiment, multiple blowers are hot-swappable. Assume all blowers are operational. If one of the multiple blowers weakens or fails (or if the strength of the airflow through the blower is too weak (or below a threshold)), the corresponding valve 107v may close due to insufficient pressure difference on both sides of the valve 107v due to gravity and / or the valve 107v closing. Valve 107v may generate an indicator signal in response to this valve closing. The problematic blower can be removed and replaced with a new unit, and the remaining blowers continue to operate.

[0150] In one embodiment, several blowers can be operated, while others can be kept idle as "backups" during normal operation. If one of the operating blowers is detected to have "failed" or "weakened," a backup blower will start up and take over the role of the "active" blower.

[0151] In the cooling system shown in Figure 6 (e.g., 20), the airflow from the refrigerant chamber CSC to the condenser 108 depends on the blower 107, but is not limited to this. The airflow from the refrigerant chamber CSC to the condenser 108 can be generated naturally without using the blower 107, by the vapor pressure difference between the refrigerant chamber CSC and the space between the output of the condenser 108 and the surface of the liquid refrigerant in the tank 102.

[0152] Figure 18 is a schematic diagram of cooling systems 30a and 30b according to embodiments of the present invention. Unlike the cooling system 20 in Figure 6, cooling systems 30a / 30b do not include a blower. Instead, cooling systems 30a / 30b include tubes 31a / 31b connected between the refrigerant chamber CSC and the condenser 108. In cooling system 30a, tube 31a is connected to the top of the refrigerant chamber CSC, while in cooling system 30b, tube 31b is connected to the bottom of the refrigerant chamber CSC.

[0153] Furthermore, the condenser 108 shown in Figure 18 may be a heat exchanger including multiple condensation channels. An opening (117) is formed connecting the multiple condensation channels to the refrigerant tank 102, thereby allowing the condensed liquid coolant to drip out and be collected by the refrigerant tank 102.

[0154] The evaporation space VPS represents the space within the refrigerant chamber CSC, the condensation space CDS represents the space within the condenser 108, and the post-condenser space ACS represents the space between the output of the condenser 108 and the refrigerant liquid level in the refrigerant tank 102. Since these spaces are not inherently shielded, the temperature and vapor pressure in VPS and ACS are approximately uniform but should differ from each other. On the other hand, the temperature and vapor pressure in CDS differ between the inlet of VPS and the outlet of ACS. In Figure 18, the evaporation space VPS and the condensation space CDS are sealed except for the connection ports. The evaporation space VPS and the condensation space CDS are separated / isolated from each other. Figure 18 also shows the evaporation space VPS, the condensation space CDS, and the post-condenser space ACS.

[0155] In Figure 18, the pump 105 connecting the refrigerant chamber CSC and the refrigerant reservoir / tank 102 is omitted. Instead, a pump 105' is shown that sends liquid refrigerant collected from the refrigerant collector at the bottom of the CSC to the refrigerant injector at the top of the refrigerant chamber CSC, thereby facilitating the circulation of liquid refrigerant within the CSC.

[0156] The purpose of Figure 18 is to address two types of connections between the refrigerant chamber CSC (bottom vs. top) and the condenser 108. This is to accommodate the needs of different operating modes (for example, when the refrigerant vapor is heavier than air, it is connected as 30b in the normal operation and shutdown sequence PDS, and as 30a in the power-up sequence PUS). Only a portion of the refrigerant chamber CSC is shown in the figure.

[0157] During the cooling operation, heat-generating components (e.g., CPU or GPU) continue to operate normally and generate heat. The first pressure P corresponds to the evaporation zone within the evaporation space VPS or refrigerant chamber CSC. VPS This corresponds to the second pressure P in the space after the condenser ACS. ACS It becomes higher than. The airflow carrying the mixture of refrigerant vapor and air has a vapor pressure difference ΔP=P between the first pressure in VPS and the second pressure in ACS. VPS -P ACS This is generated by the following: Here, the evaporation zone VPS is a subspace of the evaporation space, where the liquid refrigerant evaporates as refrigerant vapor. Similarly, the post-condenser space ACS is a subspace where most of the refrigerant vapor condenses into liquid refrigerant as it passes through the condenser 108, and the temperature in ACS is the same as the refrigerant in tank 102, which may be 3-8°C higher than the supplied service fluid.

[0158] Vapor pressure P VPR Since this is directly related to the vapor temperature, as long as the majority (more than 90%) of the mixed gas in the VPS, CDS, and ACS is refrigerant vapor (for example, more than 90% of the ambient air is removed during the air purge (degassing) phase in the startup sequence PUS, as shown in Figure 24), the operation described in the previous paragraph can be rephrased as follows: During the cooling operation, heat-generating components (e.g., CPU or GPU) continue to operate normally and generate heat. The first temperature T corresponding to the evaporation zone VPS in the refrigerant chamber CSC VPS This corresponds to the second temperature T in the condenser-back space ACS within tank 102. ACS It becomes higher than . The airflow carrying the mixture of refrigerant vapor and air has a temperature difference ΔT = T between the first and second temperatures. VPS -T ACS It is generated by [the following method / system].

[0159] For example, the first temperature in the evaporation space (VPS) can be set to the boiling point of the liquid refrigerant (e.g., 56°C, corresponding to the first temperature) or slightly higher, and the second temperature in the post-condenser space (ACS) can be set to 3-8°C higher than the temperature of the supply service fluid (e.g., set to 30°C from the cooling tower or 10°C from the chiller, corresponding to the second temperature, resulting in an ACS temperature of 33-38°C or 12-18°C). The vapor pressure difference corresponding to the temperature difference between the first and second temperatures is 108 Inlet A natural airflow is generated by ΔP. This naturally occurring airflow establishes refrigerant circulation via VPS, CDS, and ACS in systems 30a and 30b without the need for a blower 107, as shown in system 20 in Figure 3.

[0160] Figure 19 shows Z 103 (Z / vertical position of port 103) from Z 104 The diagram shows the temperature profiles of both the service fluid channel (solid line) and the refrigerant vapor channel (dashed line) of the condenser 108 with respect to the Z / vertical direction up to the Z / vertical position of port 104.

[0161] In Figure 19, temperature is shown on the vertical axis, and the Z-axis (vertical position) is shown on the horizontal axis. DP@108Inlet (Vertical axis) 10⁸ (10⁸) that receive refrigerant vapor from CSC Inlet This represents the dew point temperature at the inlet (also written / abbreviated as Z), 103 and Z 104 The horizontal axis represents the Z position of port 103 and port 104, respectively. The condenser 108 receives service fluid (e.g., raw water) from ports 103 / 104 and returns it to ports 103 / 104. 103 / T 104 This represents the temperature of ports 103 / 104. 103-104 The line represents the distance between ports 103 and 104 of condenser 108. The solid line represents the temperature change of the service fluid (e.g., water) in the service fluid channel of condenser 108. The dashed line represents the steam temperature change in the steam-air channel of condenser 108. 103represents the flow rate or volumetric velocity of the service fluid flowing into port 103. The thick solid / dashed curve corresponds to VV 103 for very high operating conditions. The thin solid / dashed curve corresponds to VV 103 for very low operating conditions.

[0162] When the flow velocity of the service fluid is slow / fast (refer to curve 61 / 63 labels: VV 103 is low / high), after entering port 103, since the heat dissipation per unit volume of the service fluid is high / low, the temperature of the service fluid rises more quickly (curve 61) / more slowly (curve 63) towards T DP@108Inlet and the slope of curve 61 / 63 at Z 103 increases / decreases.

[0163] VV 103 is low and the slope of the temperature profile of the service fluid at Z 103 is higher than the slope of (T DP@108Inlet -T 103 ) / L 103-104 , the temperature profile of the service fluid becomes horizontal as it approaches Z 104 and the slope becomes flat (refer to curve 61). The heat exchange between the vapor and the service fluid is positively correlated with the temperature difference ΔT VP between the refrigerant vapor T FF and the service fluid T VP-FF . Therefore, the rapid decrease in ΔT 103 from Z 104 to Z VP-FF means that the gradient of the vapor temperature (refer to curve 62) flattens towards Z 104 , the temperature change near Z 104 becomes gentle, indicating that the pressure gradient at 108 Inlet is low, and as a result, the vapor flow velocity at the interface from the refrigerant chamber CSC to the condenser 108 becomes slow.

[0164] When the flow rate of the service fluid is high and the slope of the service fluid temperature profile at Z 103 is (T DP@108Inlet -T 103 ) / L 103-104When the gradient is smaller than Z, the gradient of the service fluid temperature profile is Z 104 It rises towards (see curve 63). VV 103 The temperature rises, and the service fluid temperature T at the return port 104 increases. 104 is T DP@108Inlet At significantly lower temperatures (e.g., 4-9°C), such a large temperature difference ΔT between the refrigerant vapor and the service fluid occurs. VP-FF This facilitates heat exchange between the steam and the service fluid, and the return port Z 104 A sharp rise in the service fluid temperature in the vicinity (see curve 63) and condenser 108 Inlet Both a sharp drop in steam temperature just before the point (see curve 64) and this occur. Such a steep gradient of steam temperature change is 108 Inlet This means that the pressure gradient is high, and the vapor flow rate at the interface from the refrigerant chamber CSC to the condenser 108 is high.

[0165] As shown in Figure 19, in the case of a blowerless embodiment (e.g., cooling system 30a / 30b), the cooling efficiency is 108 Inlet The temperature difference T between the dew point temperature at output port 104 and the fluid temperature at output port 104. M , in other words, T M =T DP@108Inlet -T 104 It can be estimated by: 1) VV 103 (Service fluid, specifically the flow rate of raw water), 2) P CSC (The (sensed) pressure within the CSC, or equivalent P 108Inlet , condenser 108 inlet pressure), and 3)T 103 It can be determined by the (temperature of the raw water). Dew point temperature T DP@108Inlet The (sensed) pressure P CSC or P 108Inlet It can be obtained based on temperature T. 104 This can be obtained by temperature sensors placed around port 104. Port 104 of the condenser 108 is for returning the service fluid.

[0166] In this regard, the optimal vapor pressure difference ΔP VPR , temperature difference T M, or furthermore, in order to achieve the optimal cooling effect or cooling capacity, P CSC or P 108Inlet VV according to 103 A control method to adjust this can be proposed.

[0167] In one embodiment, the cooling system of the present invention (e.g., 30a / 30b) may include a controller 32 and a sensor 34. The controller 32 senses pressure P CSC The flow rate of the service fluid VV 103 It is configured to control or adjust the sensor 34, which senses the pressure P CSC The sensor 34 is configured to acquire the following. The sensor 34 may be located inside the refrigerant chamber CSC or at the inlet of the condenser 108.

[0168] In one embodiment, the controller 32 may be configured to execute a control scheme shown in the following pseudocode.

number

[0169] The general principle of the above control method is the sensed pressure P. CSC If the flow rate is slightly greater / less than the target pressure P0, then VV 103 Moderately increase / decrease the sensed pressure P CSC If the flow rate VV is significantly greater / less than the target pressure P0, 103 This involves significantly increasing / decreasing the value. Note that the increment / decrement values ​​1 and 4 shown in the pseudocode above are for illustrative purposes only. Those skilled in the art may modify them according to their actual requirements.

[0170] In one embodiment, P CSC When the operating range is 1.0 to 1.5 kPa (g), P0 is 1.25 kPa (g) ≈ 12.7 g / cm². 2 dP can be set to 0.25 kPa(g).

[0171] The same control scheme as above is used for T MIt can also be applied to the following. In one embodiment, the controller 32 can be configured to execute the following control scheme.

number

[0172] The same arguments as in the previous section apply here, but they will be omitted for brevity.

[0173] Furthermore, conventional two-phase immersion cooling faces another significant operational and environmental challenge: the cumbersome and dangerous task of removing server boards for service and maintenance. When servicing high-power server boards, it typically requires lifting large, heavy PCBs saturated with refrigerant from deep immersion tanks, inevitably resulting in refrigerant dripping and spilling onto the data center floor. This not only causes significant operational disruption but, more seriously, leads to the release of perfluoroalkyl and polyfluoroalkyl compounds (PFAS), raising serious environmental, health, and regulatory compliance concerns. Balancing ultra-high-density cooling with clean, safe, and environmentally friendly maintainability is a critical objective in this field.

[0174] To fundamentally resolve this dilemma, the Stop Sequence (PDS) applied to the cooling system of the present invention will fundamentally transform the maintenance experience. PDS is a multi-stage post-cooling process that functions as a pseudo-encapsulated "automatic drying / cleanup" procedure before opening the server enclosure in the cooling system of the present invention. By incorporating processes such as recirculation forced evaporation and condensation (rFEC), this Stop Sequence can actively remove more than 99% of both liquid and vapor refrigerants from component areas within the CSC that IT (information technology) technicians need to access for service. This ensures that the PCBs are substantially dry when IT technicians access them, thereby eliminating messy dripping, significantly reducing the release of regulated refrigerant vapors (PFAS) into the environment, and making it easier to slide boards horizontally for maintenance service. This groundbreaking PDS sequence redefines high-density cooling maintenance, being clean, safe, and compliant with the highest environmental protection standards.

[0175] The PDS sequence begins by stopping a full-intensity normal GPU-CPU workload (step S10) and starting a low-intensity stop workload. Subsequently, a controlled termination of the liquid refrigerant circulation (step S11) takes place, which begins by stopping the main refrigerant pump 101. As the refrigerant continues to evaporate due to the heat from the low-intensity stop workload, the recirculation pump 105 is stopped when the refrigerant level in the recirculation trench of the liquid refrigerant collector 110 falls below a predetermined "minimum" level. In this termination step, the heat from the low-intensity workload is used to convert most of the liquid refrigerant remaining in the CSC into refrigerant vapor, which is then condensed back into liquid refrigerant in the RCU, storing the liquid refrigerant in the refrigerant tank 102 and blocking the liquid refrigerant from (re)entering the refrigerant chamber CSC, thus preparing the refrigerant chamber CSC for the subsequent drying process.

[0176] Maintaining the GPU-CPU at a low-intensity workload indicates that the GPU-CPU functions as, and can be considered as, a type of heating element. The heating element in this invention is configured to vaporize the liquid refrigerant remaining in the CSC after the GPU-CPU has shut down under normal or maximum intensity. The heating element in this invention is not limited to a dummy load or a low-intensity load GPU-CPU, but can incorporate a resistive layer printed on a circuit board, or any suitable implementation, and is within the scope of this invention.

[0177] After the circulation is stopped (completion of step S11), all heating elements, including heat-generating components such as CPUs and GPUs, can be reactivated with a high-power workload to vaporize any remaining liquid refrigerant on the PCB or anywhere within the CSC (step S12). Step S12 is called the “pseudo-dryer” procedure and may involve carefully controlled execution of multiple steps to maximize the recovery of liquid and refrigerant vapor within the refrigerant chamber CSC.

[0178] In the present invention, the cooling system may include a heating element. The heating element is configured to generate heat to evaporate the liquid refrigerant, particularly after the supply of liquid refrigerant from the refrigerant tank to the CSC is stopped. The heating element may be a resistive layer (fabricated by printing or other means) placed on the main circuit board or auxiliary circuit board in the refrigerant chamber CSC, or it may include a resistive layer. The heating element may be a heat-generating component such as a CPU-GPU, SSD, HBM, PMIC, or NIC. In other words, by having the CPU-GPU execute a dummy program, the CPU-GPU and associated components may generate heat during the execution of the dummy program, and thus the computing components such as the CPU-GPU can be considered a heating element in the context of the PDS. In one embodiment, temperature sensors may be provided on the CPU-GPU, PMIC, etc., to monitor temperature changes and control the temperature in real time. A controller is also provided (which may be embedded in the CPU-GPU) to control the operation of the dummy program.

[0179] First (step S12a), at this stage, a large amount of liquid coolant remains in the immersion pocket and may be hidden in cracks under / between PCB components. Rapidly increasing the temperature can generate explosive vapors, potentially causing mechanical damage. The purpose of step S12a is to reduce this risk by gently removing most of the liquid coolant hidden in gaps and cracks (where explosive vapor generation causes the most significant damage) by raising the temperature moderately over a short period of time. In step S12a, the heating element is activated and the temperature on the silicon (or generally the temperature corresponding to the heat-generating component) is raised to T BP It is initially set to +Q℃. Here, T BP This is the boiling point of the refrigerant (for example, the refrigerant Opteon). TM In the case of 2P50, T BP Assuming that =49℃ and Q is set to 3℃, the target temperature in silicon is T BP (+Q℃ = 49 + 3 = 52℃). In one embodiment, this temperature (e.g., 52℃) is maintained for a short time, about 10 to 30 seconds, to ensure that more than 95% of the liquid coolant hidden under / inside / between the PCB components evaporates.

[0180] Second stage (step S12b): After the completion of the first step S12a, concerns about explosive vapor generation are eliminated, and it becomes possible to set the temperature even higher here to shorten the time required to vaporize all of the refrigerant remaining in the CSC. In step S12b, the heating element is activated and the temperature offset Q is set to 36°C (for example, T BP The temperature rises to the boiling point (T) of the entire space within the CSC (T) (+Q℃ = 49 + 36 = 85℃) and can be held there for 2 to 5 minutes. BP The temperature rises to 36°C higher, and all liquid refrigerant remaining in the chamber CSC is rapidly vaporized.

[0181] Steps S12a and S12b are for illustrative purposes only and can be modified according to actual circumstances. The main purpose of step S12 is to vaporize the liquid refrigerant remaining in the CSC.

[0182] After using high temperature in step S12b to promote vaporization of the liquid refrigerant in the CSC, the CSC may remain in a high vapor pressure state. If the door of the CSC is opened in this state, such high-pressure vapor may condense into a mist of liquid refrigerant and leak into the data center's HVAC system. The purpose of step S14 is to avoid or minimize this problem by first lowering the temperature of the CSC to near (or slightly higher than) the room temperature of the data center and removing the high concentration of vapor using condenser 108.

[0183] In step S14, the temperature offset Q is reduced to, for example, -22°C, T BP The temperature is set to +Q℃ = 49 - 22 = 27℃ (slightly higher than the office temperature of 22-23℃), and while maintaining this temperature for 2-3 minutes, the condenser 108 is used as a dehumidifier (more precisely, a vapor removal device) to condense / remove vapor from the mixture of vapor and air, significantly reducing the refrigerant vapor pressure in the CSC. Finally, the connection between the refrigerant chamber CSC and the ambient air (for example, via the one-way valve 703 in Figure 22 if the vapor is heavier than air) is opened, allowing ambient air to slowly flow into the refrigerant chamber CSC. As a result, residual vapor in the CSC is purged / flushed into the condenser 108 (for example, via port 712 and valve 702 if the vapor is heavier than air) as the ambient air moves, minimizing the amount of refrigerant vapor in the CSC. During step S14, the flow rate VV 103 The level is maintained at a sufficient level, and any residual vapor is condensed and stored in the refrigerant tank 102, keeping it away from IT technicians performing PCB maintenance work.

[0184] The above value for Q is for illustrative purposes only, and those skilled in the art can modify it according to actual circumstances.

[0185] Furthermore, in order to perform steps S12a, S12b, and S14, the temperature sensor can be incorporated into the refrigerant chamber CSC and placed near the heat-generating components or silicon components.

[0186] Steps S12a and S12b can be considered as a vaporization process that converts the liquid refrigerant in the CSC into vapor. Step S14 can be considered as a vapor removal process (similar to dehumidifying atmospheric humidity) that returns the refrigerant vapor in the CSC and RCU back to liquid refrigerant and stores it in the refrigerant tank. In summary, steps S12a, S12b, and S14 serve to remove all liquid or vaporized refrigerant from the refrigerant chamber CSC to the refrigerant tank in the recirculation and condensation unit RCU.

[0187] Meanwhile, in step S16, the blower 107 continuously circulates the mixture of vapor and air through or toward the condenser 108. Step S16 achieves several objectives simultaneously: 1) Condensing the refrigerant vapor using the condenser in the RCU and returning it to the liquid phase. The liquid phase refrigerant is stored in the refrigerant tank 102. 2) Assisting step S14, actively reducing the vapor pressure of the CSC refrigerant to a value close to the saturated vapor pressure at "room temperature" or ambient temperature.

[0188] Subsequently, the power to the blower 107 can be turned off (step S18). Since the blower is often used (together with the regulator 109) to generate negative pressure within the CSC (vacuum-assisted chassis sealing), turning off the power to the blower allows the negative pressure within the CSC to be equalized with the ambient air.

[0189] The steps of the shutdown sequence (using a blower) can be summarized as the post-cooling process PDS1 shown in Figure 20. The shutdown sequence and / or post-cooling process can be considered to include an evaporation (refrigerant drying) process carried out in the CSC and a subsequent de-evaporation (vapor removal) process carried out in the RCU, which are configured to remove 99% of the refrigerant (liquid or vapor) from the refrigerant chamber CSC before opening the refrigerant chamber CSC.

[0190] The chamber CSC is effectively dried and vapor-removed (dehumidified), and the pressure is equalized, allowing the front panel to be safely opened. In other words, after the execution of a post-cooling process or shutdown sequence, technicians can remove the PCB containing faulty components for maintenance by sliding it horizontally (for example, with the help of an elongation slide along the Y direction), as shown in Figure 21. Eliminating the need for a vertical lifter and being "simple, neat, and clean," the horizontal movement of the PCB made possible by this invention also enables highly efficient backplane configurations that were virtually impossible with conventional 2P-LIC vertical lift arrangements.

[0191] Simply put, the PDS (Post-Cooling) process of the present invention maximizes package density while eliminating the cumbersome and cumbersome task of lifting the board from the deep immersion tank. This prevents refrigerant dripping, allows the use of chassis designs without vertically opening sealed lids, or allows the PCB to be vertically lifted and removed from the deep immersion tank when maintenance is required. This rigorous and ordered shutdown approach ensures that high-performance computing not only maximizes the package density / efficiency of GPU-CPUs but also combines it with an operationally clean, environmentally conscious, safe, and efficient maintenance model.

[0192] Importantly, this sequence represents a fundamental breakthrough in minimizing the environmental impact of the 2P-LIC system. The forced two-stage regeneration process consists of powerful "pseudo-dryer" vaporization (S12a-S12b), followed by dehumidification of condenser 108 (S14), and a continuous vapor circulation cycle (S16), ensuring near-complete refrigerant recovery. By removing almost all residual liquid refrigerant and refrigerant vapor from the refrigerant chamber CSC before opening the front panel, this method significantly reduces leakage discharge. This ultra-efficient containment is essential for significantly reducing the global warming potential (GWP) and proactively mitigating the risk of PFAS leakage and environmental pollution associated with maintenance procedures.

[0193] It should be noted that the post-cooling process PDS1 in Figure 20 is for a cooling system with a blower. Furthermore, the present invention also provides a post-cooling process or shutdown sequence for blowerless configurations (e.g., cooling systems 30a / 30b shown in Figure 19). Similarly, the shutdown sequence for blowerless configurations is initiated to ensure that the refrigerant is recovered almost completely into the reservoir, minimizing the potential leakage of volatile organic compounds (PFAS) and their environmental impact.

[0194] Figure 23 shows process PDS2 (PDS in a blowerless configuration). Process PDS2 begins by stopping the supply of liquid refrigerant from the refrigerant reservoir in the RCU to the refrigerant chamber CSC (step S20). This is followed by a controlled refrigerant drying phase (or (pre) drying process, step S22, which is a process that gently expels any liquid refrigerant hidden under or between the components and the PCB before initiating the forced main drying process). At this point, the internal heating element or CPU / GPU is activated, and the CSC temperature T CSC The vapor pressure P in the refrigerant chamber CSC is raised to 3-5°C above the boiling point of the refrigerant (at 1 ATM) and maintained there (step S22a). This operation raises the vapor pressure P in the refrigerant chamber CSC. VPR The pressure rises to 1.05-1.3 ATM at the condenser inlet, creating a high-pressure gradient that drives the steam flow across the condenser. Simultaneously, the flow rate of raw water (service fluid) into the condenser increases VV 103 The speed increases and is maintained at a large speed, for example, 50-70% of the full speed (step S22b). This results in a large T M (See Figure 19) is formed, and 108 inlet A high vapor pressure gradient is formed, and the vapor pressure P inside the refrigerant tank VPR This will drop significantly to approximately 0.2-0.4 ATM. 108 Inlet The high vapor pressure gradient in is essential for the movement of refrigerant vapor from the CSC into / through the condenser, while the low vapor pressure in the ACS within the refrigerant tank, combined with gravity, is essential for air-vapor movement and the fall of condensed refrigerant into the refrigerant tank 102.

[0195] The drying process involves the pressure difference ΔP in the condenser.108.In-out =(P 108Inlet -P 108.Outlet ) to track or the temperature difference ΔT of the service fluid 104-103 =( T 104 -T 103 This can be monitored by tracking P. 108Inlet / P 108.Outlet represents the pressure at the inlet / outlet of the condenser, T 103 / T 104 This represents the temperature of the raw water port / return port that receives / returns the raw water (or generally the service fluid) to the condenser. When the rate of change of these indicators falls below a predetermined threshold (e.g., less than 30 Pa / sec or less than 0.06 °C / sec), it means that the rate at which vapor condenses in the condenser (grams / sec) is approaching zero. In other words, it means that there is very little vapor left to condense in the refrigerant chamber CSC. In other words, the CSC is essentially dry, the system is considered to have reached equilibrium, and the vapor pressure at all locations is P Satu@T103 (P VPR ≒P Satu@T103 ) becomes close to this. Here, P Satu@T103 is temperature T 103 This represents the saturated vapor pressure at the temperature of the source water or the service fluid that is typically flowing in.

[0196] If the CSC and RCU remain closed at this point, the pressure inside the CSC and RCU will be P AIR +P VPR This is the result. Here, P AIR This is the air pressure after the air purge (degassing) phase S24 of the startup sequence (details below), and P VPR is the vapor pressure. For example, in one embodiment, P AIR +P VPR =0.1ATM + 0.3ATM = 0.4ATM, or -0.6ATM(g) = -8.8PSIG (for reference, the inflation pressure for an NBA-standard basketball is 8PSIG). On the other hand, such a strong negative pressure requires thick panels, a robust structure, and sophisticated sealing, which presents challenges in terms of chassis structure and operation. However, on the other hand, as will be discussed later, it also creates an excellent opportunity for steam flushing.

[0197] The next phase / process is steam flushing (step 24). The power supply of all heating elements including the CPU / GPU is turned off (step 24a). When a valve arranged diagonally at the recirculation and condensation unit RCU connection is opened (step 24b) (details will be described later), ambient air is introduced into the coupling space between the cooling chamber CSC and the recirculation and condensation unit RCU by negative pressure. The introduced air is introduced from the opposite side in the horizontal / diagonal direction of the recirculation and condensation unit RCU connection and from the bottom (or top) of the cooling chamber CSC depending on whether the introduced refrigerant vapor is lighter (or heavier) than the ambient air. The flow rate is carefully controlled to form a smooth and sweeping airflow pattern, and the inflowing ambient air pushes the refrigerant vapor remaining in the cooling chamber CSC toward the condenser inlet 108 Inlet . During this process, the refrigerant vapor is continuously swept (ejected) from the cooling chamber CSC to the recirculation and condensation unit RCU, and a non-equilibrium state is maintained. The steam flushing ends before the airflow from the surroundings completely stops by closing the connection from the refrigerant chamber CSC to the recirculation and condensation unit RCU (such as blocking the valves 701 / 702 in Fig. 22) when the pressure P CSC in the refrigerant chamber CSC is close to the ambient pressure (for example, P Ambient - P CSC = 150 - 300 Pa or 1.5 - 3.1 g / cm 2 )(step 24c), or when the ambient air fills the refrigerant chamber. By closing the opening of 108 Inlet to separate the CSC and the RCU, most of the remaining refrigerant vapor is confined within the recirculation and condensation unit RCU, so that the vapor will not reverse-diffuse into the CSC even when the airflow from the surroundings stops. As a result, the final P VPR in the refrigerant chamber CSC becomes much lower than P Satu@T103 . Thereby, the pseudo-confinement of the refrigerant in the RCU is effectively completed, and the release of the refrigerant into the environment due to the reverse diffusion of the vapor from the RCU to the CSC when the CSC is opened for maintenance is minimized.

[0198] As discussed, P CSC while P Satu@T103 If the CSC and RCU remain closed until reaching P, the pressure inside the CSC and RCU may reach -0.6 ATM(g) or 8.8 PSIG. This is a very high pressure, making it difficult to achieve the goal of "smoothly drawing ambient air into the CSC". One solution to this problem is to overlap process steps S22 and S24 in a controlled manner. First, start the drying step S22 and wait until the pressure inside the CSC drops to a predetermined pressure level (e.g., -2 kPa or 0.3 PSIG), then start the steam flushing step S24 and execute these two process steps in parallel. In this approach, the value of VV 103 can be used to control the process. VV 103 not only affects the rate at which steam condenses into liquid in the condenser 108 (e.g., by absorbing waste heat), but also affects the T Inlet at 108 M (Figure 19), which contributes to the gradient around 108 Inlet and determines the rate of steam suction from the CSC to the RCU.

[0199] VV 103 In addition to VV, T 103 also has the potential to function as a control parameter to achieve the same effects as described above. To control T 103 , usually a chiller is required to control the temperature of the process water before supplying it to port 103 of the RCU. Considering the cost of this additional chiller, control by T 103 may not be very desirable in practice. However, nevertheless, theoretically, T 103 is a control parameter that can be implemented in the above-described hybrid S22+S24 method and is within the scope of the present invention.

[0200] There are at least two ways to position the control valve diagonally opposite the recirculation and condensation unit (RCU) connection. This depends on the relative density of the vapor and air. When the refrigerant vapor is heavier (lighter) than the air, the vapor may concentrate near the bottom (top) of the refrigerant chamber (CSC), while the air may concentrate near the top (bottom) of the CSC. In this case, when performing step 24b, it is recommended to connect the recirculation and condensation unit (RCU) to the bottom (top) of the CSC and position the valve diagonally opposite the recirculation and condensation unit (RCU) connection. This allows ambient air to gently flow into the CSC when the valve is opened during the vapor flushing step, pushing the refrigerant vapor out of the CSC into the RCU (or discharging the refrigerant vapor from the CSC) without hindering the separation of vapor and air due to density, and filling the refrigerant chamber with ambient air.

[0201] For example, Figure 22 shows a cooling system 70 according to one embodiment of the present invention. The cooling system 70 includes valves 701 to 704. When the refrigerant vapor is heavier than air, the recirculation and condensation unit RCU can be connected to the lower port 712 of the refrigerant chamber CSC with valve 702 open. In this case, the open valve 702 (or the lower port 712 of the refrigerant chamber CSC) represents the connection of the recirculation and condensation unit RCU as described in step 24b. When step 24b is performed, the diagonally positioned valve 703 is opened, allowing ambient air to flow in slowly. In this case, with the help of density-based layer separation, the lighter air remains near the top / right side of the CSC, while the heavier vapor is swept diagonally toward the RCU through the port 712 in the lower / left corner. On the other hand, when the refrigerant vapor is lighter than air, the recirculation and condensation unit RCU can be connected to the upper port 711 of the refrigerant chamber CSC with valve 701 open. When step 24b is performed, the diagonally positioned valve 704 is opened, allowing ambient air to flow in. In this case, with the help of density-based stratification, the heavier air remains near the bottom / right side of the CSC, while the heavier steam is swept diagonally towards the RCU through the port 711 in the upper / left corner.

[0202] Furthermore, the hybrid embodiment of the cooling system 70, including valves 701-704, has several advantages: 1) During the "cooling operation" phase, it is desirable to extract steam from the bottom (top) when the steam is heavier (lighter) than the ambient air; 2) Flexibility to accommodate changes in ambient air density due to site elevation, weather, etc.; 3) Flexibility to accommodate any refrigerant, regardless of whether the steam density is heavier or lighter than the ambient air.

[0203] The steps of the shutdown sequence for a blowerless configuration (operating without a blower) can be summarized as the post-cooling process PDS2 shown in Figure 23.

[0204] While the aforementioned shutdown sequence ensures clean and environmentally friendly maintenance, the cooling system also includes a startup sequence, or pre-cooling process. The startup sequence (or pre-cooling) must be performed before the server is actually started for normal operation (or before the cooling system begins its normal cooling operation).

[0205] Once installation / service is complete and the chassis is closed, the cooling chamber (CSC) is initially filled with ambient air at approximately 1 atmosphere (1 ATM). A critical challenge in the startup sequence is to purge this ambient air, which acts as a non-condensable gas, as completely as possible. If this air is not purged, the overall operating pressure (sum of refrigerant vapor pressure and residual air pressure) will become too high, significantly reducing condenser efficiency, increasing the difficulty of sealing the chassis, and hindering stable operation. Therefore, the main objective of the startup sequence is to initiate a control process that actively purges the air from the chamber. This air purging / flushing / removal process ensures that the volume (specifically the gaseous atmosphere) within the refrigerant chamber (CSC) becomes vapor-dominant, minimizing residual air pressure. Once the chamber is ready, the system can achieve and maintain the desired low-pressure operating conditions (e.g., 0.3–0.7 atmospheres) essential for stable, high-efficiency two-phase cooling within the RCU.

[0206] The startup sequence (PUS) is designed to transition the refrigerant chamber CSC and RCU from an ambient air-dominant state to an optimal vapor-dominant state. In other words, the startup sequence removes (non-condensable) air from the refrigerant chamber CSC and fills it with (condensable) refrigerant vapor.

[0207] The PUS is started (once) after it is confirmed that the front panel is securely closed and sealed (step S30). The refrigerant chamber CSC is then filled with ambient air acting as a non-condensable gas at approximately 1 atmosphere. The system control mechanism performs an initial phase (step S32) which includes sequential activation of auxiliary components: 1) The optional blower 107 (if included in the system) is activated (step S32a) and undergoes an operating range check to ensure that the required airflow and pressure difference can be maintained. 2) The main cooling pump 101 and recirculation pump 105 are activated (step S32b) to inject liquid refrigerant into the refrigerant chamber CSC and operate over their respective operating ranges. These pumps are responsible for establishing and maintaining the circulation of liquid refrigerant throughout the refrigerant chamber CSC and returning the refrigerant from the trench of the liquid refrigerant collector 110 to the reservoir 102.

[0208] Subsequently, the PUS enters the critical air purge phase (step S34), removing non-condensable air from the system, including both the refrigerant chamber CSC space and the recirculation and condensation unit RCU space.

[0209] Before proceeding, it is preferable that the connection between the refrigerant chamber CSC and the recirculation and condensation unit RCU be configured to prioritize circulating air over circulating refrigerant vapor. Referring to Figure 22, when air is lighter / heavier than vapor, the CSC-RCU is connected via ports 711 / 712 by opening valves 701 / 702 and closing valves 702 / 701.

[0210] Since the cooling system has not yet started up, its full cooling capacity is not yet available. Therefore, the GPU-CPU performs a variable-intensity startup workload instead (step S34a).

[0211] When both the main cooling pump and the recirculation pump are operating, the GPU-CPU is virtually / completely immersed in liquid coolant. The startup workload should start with a light to moderate workload. This allows the liquid coolant discharged by moderate heat to gently evaporate, and the layers of vapor and air to separate naturally, while the refrigerant vapor pressure (P) in the refrigerant chamber CSC and the recirculation and condensation unit RCU is reduced. VPR ) gradually increases.

[0212] In the blower configuration, the blower 107 draws the resulting mixture of steam and air into the condenser 108 (step S34b). In the blower-less configuration, the total pressure P TOTAL =P AIR +P VPR The resulting mixture of vapor and air is pushed into the condenser 108 (step S34b). The refrigerant vapor that enters the condenser 108 condenses as it passes through the condenser, releasing heat into the treated water, and returns to the liquid phase. The condensed liquid is recovered and returned to the reservoir.

[0213] Blower 107 or Total Pressure P TOTAL The introduction of vapor from the refrigerant chamber CSC, driven by a mechanism, followed by constant circulation and condensation of the mixture, continuously converts the refrigerant vapor back into liquid and recirculates it. Importantly, this circulation establishes a flow in which newly generated (re-condensable) vapor discharges non-condensable air from the space within the CSC chamber to the space within the RCU chamber.

[0214] In one embodiment, the refrigerant vapor generated in the refrigerant chamber CSC flows from the refrigerant chamber CSC across the condenser 108, where the recondensable refrigerant vapor is condensed into liquid refrigerant and stored in the refrigerant tank, while the non-condensable air passes through the empty volume of the refrigerant tank 102 and is discharged from the refrigerant tank 102 as an air-vapor flow A007 via valve A009 (see Figure 18). This achieves the objective of discharging residual non-condensable air trapped in the refrigerant chamber CSC and the recirculation and condensation unit RCU.

[0215] Valve A009 is opened only during the "air purge / flushing / removal" phase (step S34c) to discharge the air-vapor mixture into the atmosphere, and is closed otherwise.

[0216] The air-vapor flow A007 does not necessarily have to be discharged directly / straight into the atmosphere. For example, if server rack maintenance is infrequent, a dedicated RCU with the service fluid specially cooled to 1-2°C can be shared among multiple racks or throughout the facility to further remove evaporation of A007, thus reducing the amount of P from A007. VPR By significantly reducing emissions and ultimately releasing them into the atmosphere, the amount of refrigerant leaking from the facility can be minimized.

[0217] In blower-less configurations such as system 70 in Figure 22, there is no blower to circulate the steam-air mixture within the CSC-RCU. In such systems, the desired final operating pressure of the ACS is << 1 ATM (in the previous example, P TOTAL =P AIR +P Satu@T103 =0.1ATM + 0.3ATM = 0.4ATM), in order to expel air to the surroundings during PUS, the partial pressure P during PUS VPR Raise the P value, which is 0.2-0.3 ATM higher than the surrounding area. TOTAL.PUS It is necessary to form a target action P in the system. For example, in the system, the target action P AIR However, P AIR.OP = 0.1 ATM, and P in PUS TOTAL However, P TOTAL.PUS = 1.2 ATM, and the P required during PUS VPR PVPR.PUS =P TOTAL.PUS -P AIR.OP =1.2-0.1=1.1ATM. This P VPR.PUS This can be generated by using the heat produced by the GPU-CPU and other components on the PCB to vaporize the coolant, and since this is done under the control of a variable-intensity startup workload, it means that the purged air will be accompanied by a considerable amount of coolant vapor. For example, near the end of the "air purge" phase, during normal operation, P AIR.OP To reach 0.1 ATM, at least 9 liters of vapor would be required for every liter of purged air (!).

[0218] Therefore, in a blowerless configuration, high P during the "air purge / flushing / removal" phase VPR.PUS To maintain flow rate VV 103 It is maintained at a low level or completely shut off (or minimized or set to zero) (step S34d). This operation is the exact opposite of the blower configuration, as the degree of condensation passing through the condenser is minimized, thus reducing the high vapor partial pressure (P) generated in the refrigerant chamber CSC by the heat from the components attached to the PCB. VPR ) is properly maintained across the condenser to the refrigerant tank, with a total pressure of P at 1.2-1.3 ATM. TOTAL =P AIR +P VPR This establishes a mechanism to push air out of the condenser tank area.

[0219] Furthermore, returning to Figure 22, if the refrigerant vapor is heavier than air, it is likely that air will accumulate near the top of the refrigerant chamber CSC during the air purge phase (pushed out by the refrigerant vapor). Therefore, a port 711 located at the top of the refrigerant chamber CSC can be connected to the recirculation and condensation unit RCU of the condenser 108, and the corresponding valve 701 can be turned on (valve 702 is turned off or closed) during the air purge phase. On the other hand, during (normal) cooling operation, the refrigerant vapor is likely to accumulate near the bottom of the refrigerant chamber CSC (due to relative density and gravity). Therefore, in one embodiment, a port 712 located at the bottom of the refrigerant chamber CSC can be connected to the recirculation and condensation unit RCU of the condenser 108, and the corresponding valve 702 can be turned on (valve 701 is turned off or closed) during (normal) cooling operation. By establishing and utilizing the separation of steam and air, air can be initially sent from the CSC to the RCU, and by the time the steam finally begins to exit the RCU-CSC connection port, most of the air in the CSC has already been purged, thus in a blowerless system, the target residual air pressure level (e.g., P) can be achieved. AIR The amount of refrigerant vapor required to reach <= 0.1 ATM is significantly reduced.

[0220] Furthermore, to minimize refrigerant loss (and GWP) during the air purge phase S34, the flow resistance of regulator A009 can be set higher. This applies to both blower and blowerless configurations, but is particularly effective in blowerless configurations such as 30a and 30b in Figure 18. By selecting the CSC-RCU connection port (e.g., 711 vs 712 in the above discussion) to primarily discharge air, and then setting the flow resistance of A009 higher, P can be discharged within the ACS. TOTAL =1.2~1.3 ATM (the value required to push out non-condensable air) can be achieved without consuming a large amount of refrigerant vapor along with the A007 airflow. Alternatively, in the case of a blower-less configuration, it may be reasonable, economical, or more reasonable from a GWP standpoint to use a (portable) vacuum pump during the air purge phase instead of the above-mentioned unassisted air purge step. If such a vacuum pump is provided, the required P TOTALcan be significantly reduced (e.g., from 1.2 - 1.3 ATM to 0.3 ATM), thereby reducing the required P VPR as well (e.g., P VPR.PUS.VACUUM = P TOTAL_PURG - P AIR_Target = 0.3 - 0.1 = 0.2 ATM). This is a much lower value than the 1.1 ATM calculated previously when not using a vacuum pump, and it can simplify the starting workload during the S34 air purge phase.

[0221] This precooling startup process actively purges non - condensable air (P AIR ) from the system and continues until the volumes of the cooling chamber CSC and the recirculation and condensation unit RCU successfully transition to a vapor - dominant state, minimizing the total operating pressure (P TOTAL = P Vapor + P AIR ) in the ACS to a stable and relatively low operating range.

[0222] In a blower - less configuration, to maintain a vapor - dominant state within the system, it is most important to prevent (non - condensable) air from entering the system during normal operation and accumulating in the RCU, which would continuously increase the total pressure in the RCU and ultimately lead to system failure. One application to achieve this is to generate a slight positive pressure of 1 - 2 kPa(g) within the CSC. This restricts the refrigerant vapor from leaking slightly to the surroundings due to seal imperfection, maintaining the vapor - dominant state within the system, but if this state is not maintained, it may lead to a major failure of the entire cooling system.

[0223] In the present invention, "air purge", "air flushing", and "air removal" refer to the same concept and can be used interchangeably. Similarly, "vapor purge", "vapor flushing", and "vapor removal" also refer to the same concept and can be used interchangeably.

[0224] Innovations in structural design, particularly the absence of vertical extraction mechanisms and the integration of local recirculation and condensation units (RCUs), directly lead to a groundbreaking increase in computing density. Depending on the RCU configuration (dedicated or shared / redundant within the chassis (see Figure 2)), the number of PCBs per chassis can increase significantly. For example, in embodiments using dedicated RCUs, 12 PCBs can be arranged per chassis at a 32 mm pitch, while designs incorporating shared or redundant RCUs can accommodate up to 16 PCBs per chassis at a 29.5 mm pitch. Typically, four such chassis are arranged per rack, resulting in an extremely high density of processing power concentration, estimated at 192 to 384 GPU / CPU pairs per rack. This density is calculated as, for example, 192 pairs for 4 chassis × 12 PCBs per chassis (each chassis containing 4 processors), or 384 pairs for 4 chassis × 16 PCBs per chassis (each chassis containing 6 processors). This dramatic improvement in packaging efficiency represents a significant technological achievement that far surpasses the spatial limitations of conventional immersion systems.

[0225] This invention fundamentally solves the aforementioned shortcomings of the prior art through a novel two-phase shower immersion cooling system, achieving unparalleled packaging density, clean maintainability, and simplified infrastructure. By enabling horizontal PCB insertion / removal, vertical lifting clearance and cumbersome motorized top covers are eliminated, resulting in superior vertical packaging density and maximizing the effective use of rack space. Furthermore, horizontal access allows for short power cables and high-speed T-coders without the wiring constraints associated with V-shaped lifting clearance. BP The ability to use s optical network cables eliminates the inherent drawbacks of traditional cable wiring.

[0226] Keyly, this architecture replaces the concept of "actual immersion" with "virtual immersion," using only a fraction of the refrigerant required by conventional systems. This shower-based approach, combined with highly efficient shutdown and startup sequences, solves critical environmental and maintenance issues. Specifically, this innovative shutdown sequence incorporates a multi-stage drying process including evaporation and subsequent dehumidification, actively purging and condensing virtually all residual liquid refrigerant and vapors from the circuit board area before the chassis is opened. This ensures that the PCB is completely dry when horizontally pulled out for service, thereby preventing refrigerant dripping, significantly minimizing leak emissions, and mitigating major environmental compliance risks associated with PFAS-containing refrigerants.

[0227] To overcome the thermal limitations imposed by passive condensation and low volume utilization, the present invention uses an active blower or a vapor pressure difference (ΔP) between separate evaporation and condensation chambers. VPR By utilizing this, forced steam circulation is employed. This active steam management significantly reduces the space required for steam circulation and the size of the condenser, greatly improves the heat transfer coefficient, and increases the overall horizontal space utilization rate. Finally, this new application of active steam management, which enables adaptive control of the condenser dew point, allows for the effective raising of the required service fluid temperature, enabling the system to operate efficiently in many cases without relying on dedicated chillers or refrigerant distribution units (CDUs), thus simplifying the entire data center infrastructure.

[0228] In summary, the two-phase shower immersion cooling system of the present invention represents a fundamental paradigm shift from the limitations of conventional immersion cooling. By replacing static immersion with a gravity-driven dynamic shower, enabling horizontal service access, and utilizing an innovative refrigerant shutdown sequence, the invention simultaneously achieves several key objectives: unprecedented hardware density (up to 384 GPU / CPU pairs per rack), clean and compatible component maintainability with near-zero leakage discharge (eliminating the risk of dirty drips and PFAS), simplified cooling infrastructure by reducing refrigerant volume, eliminating the need for complex chiller / CDU components, and overcoming the thermal bottlenecks inherent in conventional IHS and D2C methods. The integration of thermal efficiency, operational cleanliness, and mechanical density provides an innovative and scalable cooling solution to meet the demands of future ultra-high-power data centers.

[0229] Those skilled in the art will readily understand that numerous changes and modifications can be made to the apparatus and method while maintaining the teachings of the present invention. Accordingly, the above disclosure should be construed as being limited only by the boundaries of the appended claims.

Claims

1. A cooling system, wherein the cooling system is Refrigerant chamber and Includes a condenser, The condenser includes a condensation channel, The cooling system is configured to execute a startup sequence before the cooling operation. The startup sequence is configured to remove air from the refrigerant chamber at least. Cooling system.

2. Includes a heating element disposed within the refrigerant chamber, A liquid refrigerant is supplied to the refrigerant chamber during the startup sequence. The cooling system according to claim 1, wherein the heating element is operated during the startup sequence, and the heat generated by the heating element vaporizes the liquid refrigerant into refrigerant vapor.

3. The cooling system according to claim 1, wherein the flow of service fluid supplied to the condenser is stopped or maintained below a threshold so that the condensation channel reaches the vapor pressure of the refrigerant chamber during the startup sequence.

4. Refrigerant tank and Includes a valve connected to the empty space of the refrigerant tank, The cooling system according to claim 1, wherein the valve is opened during the startup sequence and closed during the cooling operation.

5. The refrigerant chamber includes a first port and a second port, One of the first port and the second port is located at the top of the refrigerant chamber, and the other of the first port and the second port is located at the bottom of the refrigerant chamber. The cooling system according to claim 1, wherein the condenser is connected to the first port during the startup sequence and to the second port during the cooling operation.

6. The cooling system according to claim 1, wherein the cooling system is applied to the chassis within the server rack of the server system.

7. A pre-cooling method applied to a cooling system, wherein the pre-cooling method is The step includes executing a startup sequence before the cooling operation of the cooling system, The startup sequence is configured to remove air from at least one refrigerant chamber. The cooling system includes the refrigerant chamber, Pre-cooling method.

8. The aforementioned startup sequence is: Injecting liquid refrigerant into the refrigerant chamber, The pre-cooling method according to claim 7, comprising activating a heating element disposed in the refrigerant chamber, and vaporizing the liquid refrigerant with the heat generated by the heating element to obtain refrigerant vapor.

9. The cooling system includes a condenser with condensation channels, The pre-cooling method according to claim 7, wherein the startup sequence includes stopping the flow of service fluid supplied to the condenser or maintaining the flow of service fluid below a threshold so that the condensation channel reaches the vapor pressure of the refrigerant chamber during the startup sequence.

10. The pre-cooling method according to claim 7, wherein the startup sequence includes turning on a valve connected to a refrigerant tank.

11. The cooling system includes a condenser located outside the refrigerant chamber. The pre-cooling method according to claim 7, wherein the condenser is connected to the refrigerant chamber via a port.

12. A cooling system, wherein the cooling system is A refrigerant chamber, wherein a liquid refrigerant is injected into the refrigerant chamber, A recirculation and condensing unit including a condenser and a refrigerant reservoir, the recirculation and condensing unit being located outside the refrigerant chamber and configured to recirculate the liquid refrigerant, A circulation of liquid refrigerant is established between the refrigerant chamber and the recirculation and condensation unit. Cooling system.

13. The cooling system according to claim 12, wherein the recirculation and condensation unit includes a pump connected to the refrigerant chamber and configured to facilitate the circulation of the liquid refrigerant between the refrigerant chamber and the recirculation and condensation unit.

14. The cooling system according to claim 13, wherein the pump is connected to the upper part of the refrigerant chamber and is configured to supply the liquid refrigerant to be injected into or to be injected into the refrigerant chamber.

15. The cooling system according to claim 13, wherein the pump is connected to the bottom of the refrigerant chamber and configured to recirculate the liquid refrigerant from the refrigerant chamber.

16. The cooling system according to claim 12, wherein the recirculation and condensation unit includes a refrigerant reservoir configured to store the liquid refrigerant for circulation of the liquid refrigerant between the refrigerant chamber and the recirculation and condensation unit.

17. The cooling system according to claim 12, wherein the circulation of the liquid refrigerant includes the liquid refrigerant that flows downstream from the top of the refrigerant chamber, passes through a circuit board located inside the refrigerant chamber, exits from the bottom of the refrigerant chamber, is recirculated to a refrigerant reservoir, and is pumped back to the top of the refrigerant chamber.

18. The cooling system according to claim 12, wherein circulation of refrigerant vapor is established between the refrigerant chamber and the recirculation and condensation unit.

19. The cooling system according to claim 18, wherein the circulation of the refrigerant vapor includes vaporization of the liquid refrigerant in the refrigerant chamber, transfer to the recirculation and condensation unit outside the refrigerant chamber for condensation, and storage in a reservoir before recirculation to the refrigerant chamber.

20. The cooling system according to claim 12, wherein the refrigerant chamber is located within the chassis of a server rack.

21. A cooling system, wherein the cooling system is Refrigerant chamber and Includes a condenser, The condenser includes a condensation channel, The refrigerant vapor flows through the condensation channel of the condenser and condenses into liquid refrigerant on the inner surface of the condensation channel. Cooling system.

22. The cooling system according to claim 21, wherein the condenser includes a heat exchanger including the condensation channel.

23. The condenser includes a plate heat exchanger, The plate heat exchanger includes multiple plates, Multiple first channels and multiple second channels are formed between the multiple plates. The plurality of first channels contain the refrigerant vapor, and the plurality of second channels contain the service fluid. The cooling system according to claim 22, wherein the refrigerant vapor and the service fluid flow in opposite directions through the condenser.

24. The cooling system according to claim 22, wherein a plurality of openings are formed at the bottom of the condenser and open onto the refrigerant tank.

Images