Power-supply architecture for hybrid cooling

Abstract

First and second cooling systems cooperate to cool a power supply as it delivers power to a load. The power supply includes a first stage, a second stage, and a high-voltage bus extending therebetween to connect the first and second stages. The bus extends between the first and second cooling systems. The second cooling system, which cools the second stage, has a cooling capacity that is in excess of a cooling capacity of the first cooling system, which cools the first stage.

Description

POWER-SUPPLY ARCHITECTURE FOR HYBRID COOLINGCross Reference to Related Applications

[0001] This application claims priority to, and the benefit of the filing date of, US Provisional Application No. 63 / 624,979, filed January 25, 2024, the contents of which are hereby incorporated by reference in their entirety.Background

[0002] A typical power supply for a computer receives ac power from a power utility and transforms it into de power at a voltage level that is more appropriate for driving semiconductor circuitry. In the course of its operation, a power supply generates waste heat. This is often dissipated into the atmosphere.

[0003] The process of heat dissipation results in a component rising to a temperature that achieves thermal equilibrium. At this temperature, the rate at which heat dissipates equals the rate at which it is being produced. As a result, temperature does not change.

[0004] The actual temperature at which thermal equilibrium occurs depends on the air temperature. In cool air, a power supply reaches thermal equilibrium at a relatively low temperature. In hot air, such as in a desert climate, a power supply heats up to a somewhat higher temperature before it reaches thermal equilibrium.

[0005] In both cases, the heat-transfer rate is the same. After all, that is what “thermal equilibrium” means. However, the higher equilibrium temperature poses a significant problem for semiconductor electronics. It turns out that the number free charge carriers in a semiconductor is highly sensitive to temperature. As a result, semiconductor circuitry is surprisingly intolerant of high temperatures. Excessively high operating temperatures for long periods tend to degrade components and shorten circuit lifetime.

[0006] A known solution for use in hot climates is to cool the power supply in a liquid. This causes it to reach thermal equilibrium at a lower temperature. However, known coolants are highly corrosive. Exposure to such coolants also shortens circuit lifetime.

[0007] To make matters worse, the power supply is placed in a sealed box that is difficult to open. This makes it more difficult to repair or replace the power supply.Summary

[0008] In one aspect, the invention features a power supply and first and second cooling systems that cooperate to cool the power supply as it delivers power to a load. The power supply includes a first stage, a second stage, and a high-voltage bus extending therebetween to connect the first and second stages. The bus extends between the first and second cooling systems. The second cooling system, which cools the second stage, has a cooling capacity that is in excess of a cooling capacity of the first cooling system, which cools the first stage.

[0009] Examples of suitable loads include mining machines, hardware used for training artificial intelligence applications, including devices that rely on graphical- processing units, and servers, routers, and switches in data centers.

[0010] Embodiments include those in which the second cooling system includes a tank for holding a coolant and wherein the second stage is sealed in the tank and those in which the second cooling system includes cooling plates through which a cooling liquid flows.

[0011] Embodiments include those in which the first cooling system uses air to dissipate heat and the second cooling system uses liquid to dissipate heat and those in which the first cooling system uses liquid to dissipate heat and the second cooling system uses liquid to dissipate heat.

[0012] Some embodiments includes a rack. In such embodiments, the first stage is one of a plurality of first stages that are mounted in the rack using hot-swappable connectors and the second stage is one of a plurality of second stages that disposed in a coolant-filled tank of the second cooling system. The connector thus extends between the rack and the tank and therefore includes a section that is in neither the first cooling system nor the second cooling system.

[0013] Also among the embodiments are those in which the first stage includes connectors and handles. The connectors permit hot swapping the first stage into and out of a rack that holds additional first stages.

[0014] In some embodiments, either one or both of the cooling systems include a cooling plate through which liquid coolant flows. In such embodiments, the first stage is in contact with the cooling plate. The second cooling system also relies on a liquidcoolant, for example by using one or more cooling plates through which liquid flows or a tank filled with a liquid coolant.

[0015] In still other embodiments, the second cooling system includes a coolant-filled tank. In these embodiments, the apparatus further includes a hashboard assembly disposed in the tank. The hashboard assembly includes a hashboard and applicationspecific integrated circuits mounted on the hashboard along with the second stage. Among these are embodiments in which the application-specific integrated circuits are constituents of one or more mining machines and thus are constituents of the load.

[0016] Further embodiments include those in which the first stage includes a powerfactor correction circuit, those in which the second stage includes a dc / dc converter, those in which the first stage includes a buck downconverter, those in which the first stage includes an ac / dc converter, and those in which the power supply receives three-phase ac power.

[0017] Embodiments also include those in which the power supply is configured to supply at least half a megawatt of power in continuous operation and those in which the high-voltage bus sustains a voltage of at least three hundred volts.

[0018] Other embodiments include those in which the first stage’s average lifetime is less than that of the second stage and those in which operation of the second stage sheds more waste heat than operation of the first stage. Still other embodiments include those in which the second stage’s average lifetime is less than that of the first stage and those in which operation of the first stage sheds more waste heat than operation of the second stage.

[0019] Still other embodiments include those in which the power supply is an ac / dc power supply, those in which the power supply is a dc / dc power supply, those in which it is a dc / ac power supply, and those in which it is an ac / ac power supply.

[0020] Still other embodiments are those in which the power supply consists of only the first stage. Among these are power supplies in which the first stage consists of a power-factor correction circuit.

[0021] In some embodiments, the first and second stages are members of a plurality of first stages, among which are hot-swappable first stages, and a separate plurality of second stages. The cardinality of the plurality of first stages is greater than the cardinality of the plurality of second stages. The first and second cooling systems cool the pluralityof first stages and the plurality of second stages, respectively. As used herein, the “cardinality” of a set of elements is the number of elements in the set.

[0022] In still other embodiments, the power supply is a dual-stage power supply having a hot-swappable first stage that is mounted in a rack. Such embodiments comprise a power-supply system that comprises a plurality of power supplies, among which is the dual-stage power supply. This plurality of power supplies also comprises a single-stage power supply having a hot-swappable first stage that is mounted in the rack. This single- stage power supply lacks a second stage, like that of the dual-stage power supply. The first cooling system cools the hot-swappable first stage, just as it does the first stage of the dual-stage power supply. The power-supply system is configured to maintain power to the load upon failure of either of the first stages.

[0023] Yet other embodiments include those in which the first stage is one of a plurality of first stages of a power supply system, each of which comprises a three-phase power-factor correction circuit that comprises a step-down buck converter that outputs an average voltage of approximately four-hundred volts.

[0024] In still other embodiments, the second cooling system comprises a cooling plate through which liquid coolant flows. In such embodiments, the second stage is in contact with the cooling plate.

[0025] These and other features of the invention will be apparent from the following detailed descriptions and the accompanying drawings, in which:Description of Drawings

[0026] FIG. 1 shows a power supply having first and second stages cooled by corresponding first and second cooling systems;

[0027] FIG. 2 shows an isometric view of the power supply shown in FIG. 1 with its two cooling systems;

[0028] FIG. 3 shows an isometric front view of an embodiment of a first stage of the power supply shown in FIG. 1;

[0029] FIG. 4 shows an isometric rear view of the first stage shown in FIG. 2;

[0030] FIG. 5 shows another view of the power supply and cooling systems shown inFIG. 2;

[0031] FIG. 6 shows an exploded view of the power supply and cooling systems shown in FIGS. 2;

[0032] FIG. 7 shows an exploded view of the hashboard assembly shown in FIG. 6; and

[0033] FIG. 8 shows an alternative embodiment of the power supply shown in FIG. 2.Detailed Description

[0034] FIG. 1 shows an ac / dc power supply 10 that delivers electrical power to a load 12 and thermal power to the environment.

[0035] Examples of a load 12 include one or more servers, mining machines, machines used for training in artificial-intelligence applications, routers, and data- storage devices, and arrays of application- specific integrated circuits, such as those used in mining machines.

[0036] The power supply 10 receives ac power from an ac source 14. The ac source 14 provides the power supply 10 with a superposition of three ac waveforms phase shifted from each other by 120°.

[0037] A first stage 16 of the power supply 10 receives this ac power, uses a powerfactor correction circuit 17 to adjust phase shifts between voltage and current waveforms on each phase, and ultimately converts the ac voltage into a de voltage. The second stage 18 receives this de voltage via high-voltage bus cables 20 and reduces this incoming de voltage to an output de voltage that is more suitable for powering the load 12.

[0038] As the traveling wave from the ac source 14 encounters various components, undesirable phase shifts occur between voltage and current waveforms in all three phases. It is therefore useful for the first stage 16 to include the power-factor correction circuit 17 to correct this phase shift, thus reducing losses in power transmission.

[0039] The first stage 16 ends with a step-down buck converter 19 that provides a de voltage to the second stage 18. Since high current caused excessive heating, it is useful for the first stage 16 to operate at high voltage, thereby reducing the amount of current required for a given power level. In a typical environment, the de voltage is on the order of 600-800 volts, which the buck converter 19 reduces to approximately 360 volts. Such a voltage level is convenient for use with existing ac / dc hardware and for further voltage conversions downstream using standard components.

[0040] The second stage 18 comprises a dc / dc converter 21 that receives the high- voltage output of the first stage 16 and converts it to a lower voltage that is suitable for consumption by the load 12.

[0041] The first stage 16 includes various components that are vulnerable to gradual deterioration, particularly at high temperatures. Such components include electrolytic capacitors, which have a tendency to dry out, and transformers, and inductors, which are wound with wires that vibrate as a result of interaction between time-varying magnetic fields and currents in those wires. As a practical matter, the mean lifetime of a first stage 18 operating continuously at high power levels is quite short, typically somewhere between one-thousand and two-thousand days.

[0042] In contrast, the second stage 18, which is essentially a dc / dc converter, does not use such failure -prone components. As a result, the average operating lifetime of the second stage 18 is considerably greater than the first stage 16. In some cases, the second stage 18 even outlives the load 12.

[0043] However, although it does not require such failure -prone components, the second stage 18 tends to be considerably less efficient than the first stage 16. As a result, it tends to run significantly hotter than the first stage 16.

[0044] The foregoing difficulties are exacerbated by the load’s incessant demands for copious amounts of electrical power. For example, a single rack of mining machines or neural-network training machines of the type used in artificial-intelligence applications operating at full tilt easily consumes as much power as hundreds of room air-conditioners operating at once. In some cases, such racks will draw half a megawatt each. Power supplies 10 of the type described herein thus operate at power levels 10 that are far in excess of conventional power supplies. As a result, techniques used for design of such power supplies are typically regarded as not being reasonably pertinent to the power supplies of the type described herein.

[0045] Much heat arises from ohmic losses. Therefore, one solution is to operate the power supply 10 at high voltages, thus reducing the current required to deliver a given amount of power or by using thicker wires to reduce wire resistance.

[0046] Another solution is to increase the rate of heat transfer away from the power supply 10 by providing a heat sink to draw heat from the power supply and dissipate it into the surrounding medium. A particularly convenient medium is the surrounding atmosphere.

[0047] Using ambient air to cool the power supply 10 has two difficulties. First, in hot climates, air cooling is less effective. This tends to limit use of the power supply 10 in desert regions, where daytime temperatures routinely excess 100°F. Second, at high power, the prodigious production of waste heat easily results in reaching thermal equilibrium of the power supply 10 at a temperature that is too high for its safe operation.

[0048] It is possible to use a liquid for cooling the power supply 10, for example, by immersing the power supply 10 in a coolant. However, most coolants are corrosive to components in the power supply. An alternative method for liquid cooling involves the use of a pump that pumps cooling liquid through an arrangement of pipes.

[0049] As noted above, the power supply’s cooling requirements vary by component. In particular, the first stage 16 and the second stage 18 have different cooling requirements. It is for this reason that FIG. 1 shows respective first and second cooling systems 22, 24 for cooling the first and second stages 16, 18 respectively.

[0050] The first and second cooling systems 22, 24 have different cooling capacities. In particular, the second cooling system 24 has a greater cooling capacity than that of the first cooling system 22. In some embodiments, the first stage 16 is air cooled and the second stage is liquid cooled. In other embodiments, both the first and second stages 16, 18 are liquid cooled but by liquid-cooling systems of different cooling capacities.

[0051] FIG. 2 shows an isometric view of a rack 26 that holds several first stages 16 that are cooled by the first cooling system 22. The rack 26 features a coolant inlet 34 and a coolant outlet 36 for circulating coolant through cooling plates 38 that cool the first stages 16.

[0052] The first stages 16 are connected via the bus cables 20 to hashboard assemblies 28 on which are mounted both the load 12 and the second stage 18. The hashboard assemblies 28 are immersed in the second cooling system 24, which in this case includes a tank 32. Between the first and second stages 16, 18, the bus cables 20 pass through ambient air and are thus outside the first and second cooling systems 22, 24.

[0053] FIG. 3 and FIG. 4 shows isometric front and back views of one of the first stages 16 shown in FIG. 2. The first stage 16 includes handles 40 for extracting the first stage 16 from the rack 26 and first hot-swap connectors 42 that engage corresponding second hot-swap connectors 44 in the rack 26 so that the first stage 16 can conveniently be pulled out and replaced.

[0054] FIG. 5 shows a rear view of the embodiment shown in FIG. 2 in which the bus cables 20 can be seen passing between the first and second cooling systems 22, 24 to connect the first and second stages 16, 18.

[0055] FIG. 6 shows an exploded view to reveal a high-voltage busbar 48 to which the bus cables 20 connect and a backboard 50 on which are mounted the second hot-swap connectors 44 that mate with the first hot-swap connectors 42. One of the first stages 16 is shown in the process of being inserted into the rack 26. In addition, one of the hashboard assemblies 28 is shown as having been extracted from the tank 32 to reveal two second stages 18 mounted to the hashboard assembly 28 with third hot-swap connectors 46 that connect to a corresponding backboard (not shown) in the tank 32.

[0056] FIG. 7 shows an exploded view of the hashboard assembly 28 to reveal a first heat sink 52 and a second heat sink 54 with a hashboard 56 sandwiched therebetween. The two second stages 18 mount to the hashboard 56 via the third hot-swap connectors 46. The load 12 is likewise mounted on the hashboard 56. In this embodiment, the load 12 comprises an array of application-specific integrated circuits 60 of the type that are often used in mining machines.

[0057] FIG. 8 shows an alternative embodiment in which the second cooling system 24 comprises cooling plates 38 similar to those used in the first cooling system 22.

[0058] The embodiment shown in FIGS. 2, 5, 6, and 8 features a power supply system that comprises several power supplies 10. In particular, there are more power supplies 10 than are necessary to actually power the load 12. These spare power supplies 10 are present to take over in case a first stage 16 fails during operation. This means that the load 12 can continue to operate while the failed first stage 16 is replaced. Since the first stages 16 are all hot swappable, replacement of the failed first stage 16 does not require shutting down power to the load 12.

[0059] In some embodiments, the number of second stages 18 matches the number of first stages 16. However, since second stages 18 fail far less often than first stages 16, the redundancy of second stages 18 is of limited value.

[0060] Accordingly, in some embodiments of the power supply system, the number of first stages 16 exceeds the number of second stages 18. Such embodiments can be viewed as having power supplies 10 that fall into two categories: those that comprise both a first stage 16 and a second stage 18 and those that comprise the first stage 16 but omitthe second stage 18. The former are considered to be “dual-stage power supplies” whereas the latter are considered to be “single-stage power supplies.”

[0061] The embodiments shown in FIGS. 2, 5, 6, and 8 feature plural first stages 16 essentially working in parallel to supply power to the load 12. In a preferred embodiment, each first stage 16 features a power-factor correction circuit 17.

[0062] Within a first stage 16, the power-factor correction supports high time-varying currents that generate time- varying electromagnetic fields. These time-varying electromagnetic fields, in turn, induce time-varying currents in the power-factor correction circuits 17 of nearby first stages 16. The resulting coupling between the first stages 16 leads to imbalances that make it difficult for the first stages 16 to cooperate in parallel to power the load 12.

[0063] One approach to addressing this difficulty is to provide physical shields between first stages 16. However, such shields are bulky and add cost.

[0064] An alternative solution is to dispense with shielding altogether and to instead provide each first stage 16 with three separate inputs, one for each of the phase-shifted ac waveforms provided by the ac source 14.

[0065] In such an embodiment, each first stage 16 will be handling three times the current of a first stage 16 that receives only one of the three phase-shifted ac waveforms. This obviously results in greater electromagnetic interference between the first stages 16, particularly given their proximity to each other within the rack 26.

[0066] However, because each first stage 16 receives the same set of three phase- shifted ac waveforms from the same ac source 14, this hardly matters. Even though the first stages 16 are poorly isolated from each other, both because of proximity and the high currents within each first stage 16, the coupling between first stages 16 is essentially identical. As a result, the first stages 16 are able to operate in parallel to efficiently provide power to the second stages 18 and, ultimately, to the load 12.

[0067] As a result, to promote achievement of effective parallel operation between first stages 16, and in particular, between power-factor correction circuits 17 in those first stages 16, it is useful for each first stage 16 to use a three-phase power-factor correction circuit and a step-down buck converter that outputs an average voltage of approximately four-hundred volts.

[0068] Having described the invention and a preferred embodiment thereof, what is claimed as new and secured by letters patent is:

Claims

CLAIMS1. An apparatus for providing power to a load, said apparatus comprising a power supply, first and second cooling systems that cool said power supply while said power supply delivers power to said load, and a high-voltage bus that extends between said first and second cooling systems, wherein said second cooling system has a cooling capacity that is in excess of a cooling capacity of said first cooling system, wherein said power supply comprises a first stage and a second stage, wherein said high-voltage bus connects said first and second stages, wherein said first cooling system cools said first stage and said second cooling system cools said second stage.

2. The apparatus of claim 1, wherein said second cooling system comprises a tank for holding a coolant and wherein said second stage is sealed inside said tank.

3. The apparatus of claim 1, wherein said first cooling system uses air to dissipate heat and said second cooling system uses liquid to dissipate heat.

4. The apparatus of claim 1, wherein said first cooling system uses liquid to dissipate heat and said second cooling system uses liquid to dissipate heat.

5. The apparatus of claim 1, 2, 3, or 4, further comprising a rack, wherein said first stage is one of a plurality of first stages that are mounted in said rack using hot- swappable connectors, wherein said second stage is one of a plurality of second stages that disposed in a coolant-filled tank of said second cooling system, and wherein said connector extends between said rack and said tank.

6. The apparatus of claim 1, 2, 3, or 4, wherein said first stage comprises connectors and handles, wherein said connectors permit hot swapping said first stage into and out of a rack that holds additional first stages.

7. The apparatus of claim 1, wherein said first cooling system comprises a cooling plate through which liquid coolant flows, wherein said first stage is in contact with said cooling plate, and wherein said second cooling system uses a liquid coolant.

8. The apparatus of claim 7, wherein said first stage comprises a power-factor correction circuit.

9. The apparatus of claim 1, 2, 3, or 4, wherein said first stage comprises a powerfactor correction circuit.

10. The apparatus of claim 1, 2, 3, or 4, wherein said second stage comprises a dc / dc converter.

11. The apparatus of claim 1, 2, 3, or 4, wherein said first stage comprises a buck downconverter.

12. The apparatus of claim 1, 2, 3, or 4, wherein said first stage comprises an ac / dc converter.

13. The apparatus of claim 1, 2, 3, or 4, wherein said power supply receives three- phase ac power.

14. The apparatus of claim 1, 2, 3, or 4, wherein said power supply is configured to supply at least half a megawatt of power in continuous operation.

15. The apparatus of claim 1, 2, 3, or 4, wherein said high-voltage bus sustains a voltage of at least three hundred volts.

16. The apparatus of claim 1, 2, 3, or 4, wherein said first stage has a first-stage average-lifetime, wherein said second stage has a second-stage average-lifetime, and wherein said first-stage average-lifetime is less than said second-stage average-lifetime.

17. The apparatus of claim 1, 2, 3, or 4, wherein operation of said second stage sheds more waste heat than operation of said first stage.

18. The apparatus of claim 1, 2, 3, or 4, further comprising a power supply system comprising a plurality of first stages and a plurality of second stages having respective first and second cardinalities, wherein said first stage is one of saidplurality of first stages, wherein said second stage is one of said plurality of second stages, wherein said first cooling system cools said plurality of first stages, wherein said second cooling system cools said plurality of second stages, wherein said first stages are hot swappable, and wherein said first cardinality exceeds said second cardinality.

19. The apparatus of claim 1, 2, 3, or 4, wherein said power supply is a dual-stage power supply having a hot-swappable first stage that is mounted in a rack, wherein said apparatus further comprises a power-supply system that comprises a plurality of power supplies, among which is said dual-stage power supply, wherein said plurality of power supplies comprises a single-stage power supply having a hot-swappable first stage that is mounted in said rack and lacking a second stage, wherein said first cooling system cools said hot-swappable first stage, and wherein said power-supply system is configured to maintain power to said load upon failure of either of said first stages.

20. The apparatus of claim 1, 2, 3, or 4, wherein said first stage is one of a plurality of first stages of a power supply system, wherein each of said first stages comprises a three-phase power-factor correction circuit that comprises a step-down buck converter that attempts to output an average voltage of four-hundred volts.

21. The apparatus of claim 1, 2, 3, or 4, wherein said second cooling system comprises a coolant-filled tank, wherein said apparatus further comprises a hashboard assembly disposed in said tank, said hashboard assembly comprising a hashboard and application-specific integrated circuits mounted on said hashboard, wherein said second stage is mounted on said hashboard.

22. The apparatus of claim 1, 2, 3, or 4, wherein said second cooling system comprises a cooling plate through which liquid coolant flows, wherein said second stage is in contact with said cooling plate.

Images