Temperature control and monitoring method in FT test process of memory chip

Abstract

The invention belongs to the field of semiconductor function testing, and particularly relates to a temperature control and monitoring method in the FT testing process of a storage chip. The method comprises the following steps: establishing an electrothermal double-domain coupling characteristic model to record a mapping relation among power consumption, thermal resistance and internal junction temperature; sampling transient electrical parameters in real time when the storage chip is tested, and calculating real-time power consumption; resolving the dynamic junction temperature in real time by using a virtual mapping algorithm, and constructing a digital mapping entity of the thermal state of the storage chip; the dynamic junction temperature is compared with the actually measured shell temperature, and thermal resistance network parameters are identified on line through an adaptive filtering algorithm; pre-judging junction temperature change according to the test item jump signal and generating a feed-forward compensation instruction; finally, the dynamic junction temperature serves as the main feedback quantity to drive a fluid executing mechanism to adjust the cold and hot medium mixing proportion. According to the invention, through the active prediction control of electric heating cooperation, the heat transfer lag influence is eliminated, the accurate locking of the internal junction temperature of the memory chip is realized, and the stability of the test result and the production efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor functional testing, specifically relating to a method for temperature control and monitoring during the FT test process of memory chips. Background Technology

[0002] With the rapid development of large-scale integrated circuit technology, memory chips are increasingly widely used in high-performance computing and artificial intelligence. Final testing, a crucial step in the semiconductor manufacturing process, verifies the full electrical performance of chips under controlled temperature conditions, enabling the screening of potentially defective products. High-precision temperature control is not only fundamental to simulating real-world application scenarios but also a core prerequisite for ensuring the timing accuracy and logical correctness of memory chips under harsh operating conditions.

[0003] Temperature control and monitoring during the final testing process of memory chips involves closed-loop thermal management of the test chamber and the chip itself. This technology typically utilizes external temperature control equipment to generate a constant heat flow or cooling medium, which acts on the chip under test through heat conduction, aiming to lock the chip's internal junction temperature at the target test point in real time. To monitor the test status, the system needs to integrate sensors to acquire temperature data and adjust the temperature control actuators in conjunction with commands issued by the test equipment, providing a stable thermal environment for high-power memory chips.

[0004] Existing technologies for testing high-speed memory chips generally face the problem of mismatch between macroscopic environmental temperature control and microscopic chip junction temperature, resulting in significant lag in heat transfer and difficulty in real-time response to transient thermal changes caused by drastic power consumption fluctuations. Simultaneously, temperature control and monitoring paths are often functionally disconnected, lacking a high-bandwidth closed-loop feedback mechanism, making the system exhibit low robustness when facing complex conditions such as test interface aging or contact thermal resistance drift. Traditional solutions fail to establish a deep correlation between chip electrical characteristics and heating mechanisms, and cannot predict and compensate for the nonlinear thermal stress caused by test vector switching, resulting in frequent junction temperature oscillations and inducing yield losses. Summary of the Invention

[0005] The purpose of this invention is to provide a method for temperature control and monitoring during the FT test process of memory chips, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a method for temperature control and monitoring during the FT test process of a memory chip, comprising the following specific steps: Step 1: Establish an electrothermal dual-domain coupling feature model. By pre-calibrating the thermal characteristics of different types of memory chips, obtain and record the multi-dimensional mapping relationship between transient power consumption, structural thermal resistance and internal junction temperature of memory chips under different working modes, and form a multi-dimensional feature library. Step 2: Real-time acquisition of electrical operating parameters. During the final testing of the memory chip, the transient operating current and transient operating voltage of the memory chip when running a specific test vector are detected and recorded in real time through the high-speed synchronous sampling unit. Step 3: Perform electrothermal virtual mapping calculation. Utilize the collected transient operating current and transient operating voltage, combined with the preset calculation logic, to obtain the real-time power consumption. Retrieve the pre-established multi-dimensional mapping relationship and use the embedded virtual mapping algorithm to calculate the dynamic junction temperature inside the memory chip in real time, thereby constructing a digital mapping entity of the thermal state of the memory chip. Step 4: Implement online thermal resistance network identification, compare the calculated dynamic junction temperature with the chip casing temperature collected by the physical sensor set at the bottom of the test socket in real time, and use an adaptive filtering algorithm to dynamically identify the real-time thermal resistance network parameters between the chip junction and the test environment. Step 5: Generate feedforward compensation control instructions, receive test item jump signals issued by the test machine in real time, obtain the time node of power consumption change in advance, and predict the change in junction temperature according to the electrothermal virtual mapping model to generate compensatory fluid flow adjustment instructions. Step 6: Perform closed-loop feedback regulation, with dynamic junction temperature as the main feedback quantity and chip casing temperature collected by physical sensors as the auxiliary feedback quantity. A control voltage signal is generated through a proportional-integral-derivative controller to drive the fluid actuator to adjust the mixing ratio of hot and cold media, thereby achieving precise locking of the junction temperature inside the memory chip.

[0007] Preferably, the process of establishing the electrothermal dual-domain coupling characteristic model in step 1 includes: establishing a detailed thermal model of the memory chip using thermal simulation software, analyzing the thermal diffusion path of the memory chip under different packaging materials, pin arrangements, and different test base pressures; applying a step-by-step power load to the sample chip in an experimental environment, recording the chip shell temperature and the readings of the internal temperature sensor at different ambient temperature points, and obtaining the nonlinear compensation coefficient of thermal resistance changing with temperature; fusing the above simulation data with the measured calibration data to form a three-dimensional matrix feature table with operating voltage, operating current, and ambient temperature as input variables and junction temperature as the output variable, and storing it in the non-volatile memory unit of the temperature control system.

[0008] Preferably, the process of obtaining electrical operating parameters in step 2 includes: capturing the current waveform of the power supply pin of the memory chip at a preset sampling frequency using a high-bandwidth current monitoring circuit integrated in the power supply channel of the test instrument; simultaneously monitoring the input voltage of the memory chip using a high-speed voltage comparator to ensure the stability of the support voltage signal; the sampling frequency must be highly synchronized with the clock cycle of the test instrument to ensure that the electrical characteristics corresponding to each test vector can be completely captured, avoiding misalignment of the electrothermal relationship due to sampling delay.

[0009] Preferably, the process of performing the electrothermal virtual mapping calculation in step 3 includes: firstly, multiplying the real-time acquired transient operating current and transient operating voltage to obtain the total transient power consumption of the memory chip in the current test cycle; then, according to the current logic operation state of the memory chip, retrieving the corresponding thermal resistance reference value from the feature library; using the product of the total transient power consumption and the thermal resistance reference value as the temperature increment, superimposing it on the reference temperature measured by the physical sensor to obtain the initial estimate of the dynamic junction temperature; further, using the thermal capacity correction factor to perform time-domain smoothing processing on the initial estimate of the dynamic junction temperature to compensate for the phase shift caused by the thermal inertia of the packaging material, obtaining a high-precision real-time value of the dynamic junction temperature.

[0010] Preferably, the process of implementing online thermal resistance network identification in step 4 includes: constructing a thermal resistance observer based on state-space equations, using dynamic junction temperature as a state variable and chip casing temperature as an observation variable; using the Kalman filter algorithm to perform prediction and update steps in each sampling period; in the prediction step, calculating the predicted value at the current moment based on the thermal resistance state at the previous moment; in the update step, calculating the residual between the predicted casing temperature and the actual measured casing temperature, and correcting the thermal resistance state in real time according to the Kalman gain; the identification process can identify thermal resistance drift caused by aging of the test socket metal probe, fluctuation of contact pressure, or change of test environment humidity in real time, and feed the corrected thermal resistance parameters back to the calculation process in step 3.

[0011] Preferably, the process of generating feedforward compensation control instructions in step 5 includes: the test instrument sends a pre-trigger pulse signal to the temperature control system during a specific clock cycle before executing the high-power test item jump; after recognizing the pre-trigger pulse signal, the temperature control system extracts the expected power consumption level of the next stage from the test vector sequence; based on the difference between the expected power consumption level and the current power consumption level, the possible fluctuation amplitude of the junction temperature is calculated in combination with the electrothermal virtual mapping model; the required compensation flow rate of the fluid actuator is deduced in reverse using the fluctuation amplitude, and the opening of the hot and cold fluid proportional valve is adjusted in advance to offset the transient changes in the heat generation of the chip through advance control.

[0012] Preferably, the closed-loop feedback regulation process in step 6 includes: constructing a dual-loop control architecture, wherein the inner loop is a fluid flow control loop and the outer loop is a temperature compensation control loop; the proportional coefficient, integral time constant, and derivative time constant of the proportional-integral-derivative controller are dynamically adjusted based on the real-time thermal resistance value output by the thermal resistance network online identification module; when the dynamic junction temperature deviates from the preset target value, the controller outputs control commands to the high-precision proportional servo valve to adjust the mixing ratio of the low-temperature fluid and the high-temperature fluid; through comprehensive regulation of fluid flow rate and temperature, it is ensured that the fluctuation range of the junction temperature inside the memory chip during the entire testing process is within a predetermined threshold.

[0013] Preferably, the temperature control and monitoring method for the FT test process of the memory chip also involves real-time monitoring of the test environment, including continuous monitoring of the air flow rate, ambient humidity, and test socket contact pressure in the test chamber; when the ambient humidity exceeds a predetermined range, the system automatically starts a nitrogen purging device to prevent condensation during the low-temperature test stage; when the test socket contact pressure is lower than a predetermined threshold, the system issues an early warning signal to prompt the operator to check the mechanical contact status of the test socket to ensure the stability of the heat conduction path.

[0014] Preferably, the virtual mapping algorithm in step 3 further includes compensation logic for uneven heating in multiple parts inside the memory chip; according to the internal layout design of the memory chip, the chip is divided into a memory array area, a logic control area, and a peripheral interface area; according to the activation degree of different functional areas by different test vectors, different weight coefficients are assigned to calculate the local temperature distribution of each area, and the equivalent junction temperature value that has the greatest impact on overall reliability is obtained by weighted average algorithm.

[0015] Preferably, the fluid actuator in step 6 includes a circulating pump, a heating unit, a cooling unit, and a three-way proportional mixing valve; the circulating pump maintains the test medium in a constant flow circulation in a closed pipeline; the heating unit and the cooling unit respectively maintain the test medium at a preset high temperature state and a low temperature state; the three-way proportional mixing valve accurately distributes the ratio of hot and cold media entering the heat exchanger according to the control command, the heat exchanger is in close contact with the test socket, and the chip is rapidly heated or cooled by heat conduction or forced convection.

[0016] Preferably, the feedforward compensation control command also includes linkage control logic for the speed of the cooling fan of the test machine; when it is predicted that the memory chip is about to enter a continuous high power consumption mode, the system adjusts the fluid flow rate and simultaneously increases the pulse width modulation duty cycle of the cooling fan to enhance the convective heat transfer intensity on the surface of the test interface unit, thereby synergistically improving the transient response speed and steady-state control accuracy of the temperature control system.

[0017] Preferably, the electrothermal dual-domain coupling feature model is periodically updated through self-learning; the system automatically collects junction temperature fluctuation data and electrical parameter features during long-term testing, uses machine learning algorithms to identify potential model deviations, and automatically corrects key parameters in the multi-dimensional feature library during equipment idle periods, ensuring that the temperature control and monitoring method for the memory chip FT test process maintains optimal control performance throughout the entire life cycle of the equipment.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention establishes an electrothermal virtual mapping mechanism, transforming the temperature control system from passive ambient temperature regulation to active predictive control based on the chip's internal electrical behavior. Since the transmission speed of electrical signals is much higher than the speed of heat conduction, the system can detect power consumption fluctuations before junction temperature changes, eliminating the heat transfer lag problem that is difficult to avoid in traditional temperature control schemes. Experimental data shows that this method can reduce the junction temperature fluctuation of high-power memory chips to a very small range during test vector switching, improving the repeatability and accuracy of test results.

[0019] 2. This invention introduces an online thermal resistance network identification module, endowing the temperature control system with strong adaptive capabilities. In actual production processes, aging of the test socket, probe wear, or replacement of test carriers from different batches can all lead to changes in the physical characteristics of the heat transfer path. This invention, by monitoring the difference between the shell temperature and the virtual junction temperature in real time, can automatically identify and compensate for these changes in thermal parameters, eliminating the need for frequent manual calibration. This not only improves the stability of mass production testing but also shortens downtime caused by equipment maintenance.

[0020] 3. This invention achieves deep synergy between electrical characteristic testing and thermal control. By acquiring the pre-trigger signal from the testing equipment, the system constructs a composite control architecture of feedforward and feedback, solving the control problem caused by the drastic heat generation changes of high-speed memory chips when performing high-throughput read and write operations. This precise junction temperature control ensures that the electrical performance verification of memory chips is completed under the most stringent temperature conditions, closest to real-world applications, reducing yield losses due to temperature deviations and avoiding the risk of missed screenings due to excessively high junction temperatures.

[0021] 4. This invention reduces excessive reliance on complex external sensor hardware. By utilizing the electrical parameters of the memory chip itself as a virtual temperature sensor, the structural design of the test interface unit is simplified, the number of precision physical sensors is reduced, and the overall hardware cost of the test system is lowered. Simultaneously, the simplified physical structure also helps improve system reliability and reduces the possibility of test anomalies due to sensor failure.

[0022] 5. This invention improves overall testing efficiency. Because the junction temperature can be quickly and stably locked at the target point, the test program no longer needs to set redundant waiting delays to allow temperature equilibrium. This means that test cycle time is reduced, allowing more chips to be tested within the same equipment time, thus improving the production line's efficiency and overall equipment efficiency. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture according to the present invention; Figure 2This is a schematic diagram of the core principle framework of virtual mapping calculation based on the electrothermal dual-domain coupling feature according to the present invention; Figure 3 The logical flowchart for establishing the electrothermal dual-domain coupling feature model and multi-dimensional feature calibration according to the present invention is shown below. Figure 4 A flowchart illustrating the logical flow of online thermal resistance network identification and adaptive parameter correction according to the present invention. Figure 5 The logical flow diagram for generating feedforward compensation control instructions and power consumption prediction according to the present invention is shown below. Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the dual-loop closed-loop feedback regulation and the fluid actuator in this invention. Detailed Implementation

[0024] Example 1: Please refer to the appendix Figure 1 To be continued Figure 6 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0025] The temperature control and monitoring method for the FT test process of the memory chip performs the following specific steps: In step 1, the process of establishing the electrothermal dual-domain coupling characteristic model is specifically manifested as follows: First, a refined three-dimensional thermal conduction geometric model of the memory chip is constructed using high-precision thermal simulation analysis software. During this model establishment process, the silicon wafer, lead frame, molding resin, solder ball array, and their associated test sockets and test carriers within the memory chip are all discretized into millions of finite element mesh elements. For different packaging materials, their corresponding thermal conductivity, specific heat capacity, coefficient of thermal expansion, and density parameters are extracted from the material property library, and the nonlinear characteristics of these parameters changing with temperature are considered. For example, for silicon, its thermal conductivity is set as a function that gradually decreases with increasing temperature.

[0026] Furthermore, the thermal diffusion path of the memory chip under different packaging material combinations was analyzed. During the pin layout analysis phase, the physical location of each power and ground pin was extracted, and the parasitic resistance heating of each pin during high-speed signal transmission was calculated and applied as a local heat source to the model. Simultaneously, the contact condition between the test base and the chip surface under different pressure gradients was simulated to establish a contact thermal resistance model. By changing the contact pressure value in the simulation environment, the negative correlation mapping curve between contact thermal resistance and mechanical pressure was obtained.

[0027] In the experimental calibration phase, a sample chip was placed in a controlled constant-temperature test chamber. A stepped power load was applied to the chip via the power module of the testing equipment, with power step levels set from 50 milliwatts to 500 milliwatts. During each power stabilization phase, a high-resolution infrared thermal imager was used to capture the temperature distribution on the chip surface, and the readings from the integrated temperature sensor inside the chip were simultaneously read. The measured casing temperature, the internal sensor temperature, and the junction temperature calculated from the simulation were used for three-way verification. Through residual analysis of multiple sets of experimental data, a nonlinear compensation coefficient for the thermal resistance change with temperature was extracted. This coefficient was used to correct deviations in the idealized parameters of the simulation model.

[0028] Finally, the multi-dimensional simulation data and measured calibration data were deeply integrated to construct a three-dimensional matrix feature table with operating voltage, operating current, and ambient reference temperature as independent variables and real-time junction temperature as the dependent variable. This feature table not only covers steady-state thermal resistance parameters but also includes the heat capacity time constant reflecting transient thermal response characteristics. This three-dimensional matrix feature table is formatted as a binary data stream and stored in the non-volatile memory inside the temperature control system controller to enable sub-microsecond-level querying and retrieval during real-time control loops.

[0029] In step 2, the process of acquiring electrical operating parameters in real time specifically includes: the temperature control system and the power supply unit of the testing machine achieving hardware-level interconnection. Within the power supply channel of each memory chip, a precision current monitoring circuit with high bandwidth response and low insertion loss is connected in series. This monitoring circuit uses a Hall effect sensor or a precision sampling resistor in conjunction with a high-speed instrumentation amplifier, and its analog bandwidth is designed to be no less than 10 MHz.

[0030] The high-speed synchronous sampling unit performs analog-to-digital conversion on the analog signal output by the current monitoring circuit at a preset sampling frequency. The sampling frequency is set to 4 to 8 times the clock frequency of the test vector on the test equipment to ensure that at least 4 sampling points can be captured within the test vector period, accurately reconstructing the transient waveform characteristics of the current. At the same time, the input voltage pins of the memory chip are monitored in parallel through a high-speed voltage comparator array to acquire minute voltage ripple changes in real time.

[0031] The temperature control system's sampling control logic is synchronized at the hardware level with the test instrument's trigger signal. Whenever the test instrument issues a start trigger pulse, the synchronous sampling unit immediately initiates a burst sampling mode. The sampled operating current and voltage data are cached in real-time in a high-speed random access memory. The effective value, peak value, and average value are calculated through feature extraction of the waveform data. To eliminate electromagnetic interference in the test environment, the sampled data undergoes a first-order hysteresis filter before entering the calculation logic to remove high-frequency pulse noise, ensuring the stability and authenticity of the support signal and avoiding electrothermal mapping deviations caused by sampling data jitter.

[0032] In step 3, the process of performing the electrothermal virtual mapping calculation is specifically manifested as follows: The core processor of the temperature control system first reads the transient operating current and transient operating voltage values ​​from the synchronous sampling unit. Through the digital multiplication unit, the product of these two values ​​is calculated to obtain the total transient power consumption of the memory chip during the current sampling period. Subsequently, the processor parses the task status word from the test instrument to identify the current logical operation mode of the memory chip, such as standby mode, continuous read operation mode, page programming operation mode, or full-chip erase mode.

[0033] Based on the identified operating mode, the corresponding thermal resistance reference value is retrieved in real time from the feature library in non-volatile memory. The current temperature increment contribution value is obtained by multiplying the transient total power consumption by the thermal resistance reference value. This temperature increment contribution value represents the increase in junction temperature relative to the reference point caused by heat generated by internal circuitry activity.

[0034] To further improve accuracy, this temperature increment is superimposed on the reference temperature measured by a physical sensor to obtain an initial estimate of the dynamic junction temperature. Considering the thermal inertia of the chip packaging material, i.e., the change in power consumption and the change in junction temperature have a phase lag on the time axis, the system introduces a thermal capacity correction factor.

[0035] In step 3, to address the phase shift caused by the thermal inertia of the encapsulation material, this invention employs a time-domain smoothing compensation algorithm based on a first-order RC thermal network model. Specifically, a time-domain smoothing compensation algorithm based on the encapsulation thermal time constant is first defined. Related attenuation coefficient The calculation formula is as follows: ; in, This indicates the control cycle (sampling time interval) of the temperature control system. The thermal time constant of the memory chip package structure under test is obtained through the thermal characteristic calibration experiment in step 1. Subsequently, this attenuation coefficient is used to perform a first-order low-pass filtering on the initial estimate of the dynamic junction temperature to obtain the compensated, high-precision dynamic junction temperature. : ; in, This represents the initial estimate of the dynamic junction temperature at the current moment, calculated by multiplying the transient total power consumption by the reference thermal resistance value. This represents the dynamic junction temperature value after smoothing at the previous moment. Using this recursive formula, the system can effectively simulate the accumulation and dissipation of heat, compensating for the phase delay caused by heat conduction hysteresis.

[0036] The correction factor is based on a first-order thermal network model for time-domain smoothing. It compensates for the phase shift by calculating the difference between the initial estimate at the current time and the dynamic junction temperature at the previous time, and multiplying it by an attenuation coefficient related to the package thermal time constant.

[0037] Furthermore, the virtual mapping algorithm also incorporates compensation logic for uneven heating in multiple parts of the memory chip. In step 3, the compensation logic for uneven heating within the memory chip is as follows: First, based on the chip's physical layout design, the chip is divided into... Each functional area is defined as an independent functional region, and a weight coefficient vector is established. This weighting coefficient is not a static value, but is dynamically calculated in real time based on the type of test vector currently being executed.

[0038] Specifically, the system identifies the currently activated combination of functional areas by parsing the test vector commands sent by the test equipment. For a specific test mode... The corresponding weighting coefficient An activator strongly correlated with the test mode Decide: ; in, It is a value between 0 and 1, representing the test mode. For the The activation level of each functional area. For example, during a sequential read operation, the corresponding area of ​​the storage array... The logic control area and the peripheral interface area are respectively set as and Conversely, when performing interface calibration testing, the opposite applies.

[0039] Subsequently, the system calculates the local temperature increment for each functional region. This increment is determined by the current transient total power consumption. Each region has its own unique thermal resistance reference value. And the corresponding weighting coefficients together determine: ; Ultimately, the equivalent dynamic junction temperature Instead of a simple weighted average, a weighted root mean square algorithm is used to calculate the result, taking into account the differences in the impact of different regions on the overall reliability of the chip. ; in, The reference temperature is used (such as the casing temperature measured by a physical sensor). This algorithm can more accurately reflect the hot spot temperature inside the chip under the coupling effect of multiple heat sources, providing more accurate feedback for subsequent closed-loop control.

[0040] Based on the chip's internal layout, the model is divided into a memory array area, a logic control area, and a peripheral interface area. The processor determines the activation intensity of each functional area based on the type of test vector being executed. For example, during large-scale parallel read tests, the heat dissipation weighting coefficient of the memory array area is increased; during high-speed serial interface calibration tests, the heat dissipation weighting coefficient of the peripheral interface area is increased. A weighted average algorithm is used to synthesize the local temperature contributions of each area, calculating the real-time equivalent dynamic junction temperature value that has the greatest impact on overall reliability.

[0041] In step 4, the process of implementing online thermal resistance network identification is as follows: the system constructs a thermal resistance observer based on state-space equations. In this observer, the dynamic junction temperature is set as the system state variable, and the chip casing temperature is set as the observed variable. By introducing a Kalman filter algorithm, prediction and update steps are performed within each fixed control cycle.

[0042] In step 4, the specific implementation of constructing the thermal resistance observer based on Kalman filtering is as follows: First, the heat transfer path is represented as a first-order state-space model. State variables are defined. That is, real-time thermal resistance; observed value. That is, the chip casing temperature measured by physical sensors; control input. This refers to real-time power consumption. The system's state equation and observation equation can be expressed as: ; ; in, This is the state transition matrix, which is usually set to 0 because thermal resistance changes slowly. ; The process noise has a covariance of ; To observe the noise, its covariance is ; The observation matrix, whose physical meaning is to convert thermal resistance into a function of temperature, is as follows: ,in This refers to the dynamic junction temperature calculated in step 3.

[0043] The recursive process of Kalman filtering is as follows: Prediction steps: ; ; Update steps: ; ; ; in, This refers to the real-time thermal resistance network parameters corrected by the Kalman filter algorithm. These parameters are fed back to the electrothermal virtual mapping calculation in step 3 to update the thermal resistance reference value and form a closed-loop correction.

[0044] During the prediction phase, the optimal thermal resistance state value obtained at the previous moment and the system's heat transfer state transition matrix are used to calculate the predicted thermal resistance value and the corresponding predicted covariance matrix value at the current moment. During the update phase, physical sensors collect real-time chip casing temperature data located at the bottom of the test mount. This measured value is subtracted from the predicted casing temperature calculated based on the predicted junction temperature and the current thermal resistance model to obtain the observation residual.

[0045] The Kalman gain calculation unit calculates the optimal gain coefficient based on the current state prediction covariance and the observation noise covariance. This gain coefficient is then used to weight the observation residuals, correcting the predicted thermal resistance state. This identification process possesses extremely high sensitivity, capable of capturing even minute physical changes along the heat transfer path. For example, when the metal probe of the test fixture experiences fatigue aging due to repeated pressing, leading to increased contact thermal resistance, the observation residuals will exhibit a trend shift. The Kalman filtering algorithm can identify this change and update the thermal resistance parameters accordingly.

[0046] Similarly, if the humidity of the test environment fluctuates drastically or the mechanical contact pressure of the test seat decreases due to unstable cylinder pressure, the system can capture and correct the model in real time through the online identification module. The corrected real-time thermal resistance parameters will be fed back to the virtual mapping calculation in step 3 as the thermal resistance reference value for the next cycle, forming a closed-loop self-calibration mechanism at the parameter level.

[0047] In step 5, the process of generating feedforward compensation control commands is specifically manifested as follows: the temperature control system connects to the sequence controller of the test equipment via a dedicated hardware communication link. When the test equipment plans to execute a high-power test item jump after several subsequent clock cycles, it sends a pre-trigger pulse signal to the temperature control system in advance.

[0048] After recognizing the pre-trigger signal, the protocol parsing module inside the temperature control system extracts the expected power consumption level for the next stage from the pre-stored test sequence description file. The processor calculates the numerical difference between the expected power consumption level and the current real-time power consumption level. Combining this with the electrothermal virtual mapping model established in step 1, the predicted value of the junction temperature fluctuation amplitude caused by this power consumption jump is pre-calculated.

[0049] Using the predicted fluctuation amplitude and the inversion algorithm of the heat conduction model, the additional cooling or heating capacity required by the fluid actuator is calculated. Based on the predicted temperature rise gradient, the system calculates the pre-adjusted opening increment of the proportional valve for the hot and cold fluids. This increment command is converted into a high-frequency pulse width modulation signal to drive the servo motor of the proportional valve.

[0050] Through this proactive control mechanism, the temperature control system adjusts the fluid flow rate and mixing ratio on the heat exchanger surface before the actual power consumption of the chip changes abruptly. This event-driven feedforward control can offset the physical delay in heat transfer, allowing the heat generated by the chip to be removed or replenished in real time, ensuring that the junction temperature remains stable even during periods of dramatic power consumption changes, and eliminating the overshoot or undershoot phenomena commonly found in traditional control systems.

[0051] In step 6, the closed-loop feedback regulation process specifically includes: the system adopts a dual-loop collaborative control architecture. The inner loop is set as a fluid flow and mixing ratio control loop, whose task is to ensure that the fluid temperature and flow rate output by the actuator quickly reach the target values ​​set by the feedforward command. The outer loop is set as a junction temperature dynamic compensation control loop, using the high-precision dynamic junction temperature calculated in step 3 as the main feedback signal.

[0052] In step 6, when performing closed-loop feedback regulation, the parameters of the PID controller are not fixed, but are based on the real-time thermal resistance value output by the online thermal resistance network identification module. Dynamic gain scheduling is implemented. Specifically, the system pre-sets a parameter scheduling table to divide the thermal resistance values ​​into... There are several intervals, each corresponding to a set of optimal PID control parameters.

[0053] In real-time control, the system first obtains the current real-time thermal resistance value. Then, a linear interpolation algorithm is used to dynamically calculate the PID parameters at the current time step: ; ; ; in, , , It is the reference thermal resistance value The basic PID parameters are as follows; The gain regulation factor, calibrated experimentally, has the physical meaning that as thermal resistance increases (contact deterioration), the proportional gain needs to be increased. ) and differential gain ( To improve system response speed while reducing integral time ( This is to prevent integral saturation, thereby ensuring the stability of the system under various operating conditions.

[0054] The outer loop junction temperature dynamic compensation control loop adopts a positional PID algorithm, and its output quantity... (i.e., the correction signal) is: ; in, To set the deviation between the target junction temperature and the real-time equivalent junction temperature, This is for control period. This correction signal It is superimposed on the input end of the inner ring fluid flow control ring, driving the fluid actuator to precisely adjust the mixing ratio of the hot and cold media.

[0055] The proportional-integral-derivative (PID) controller receives the target temperature setpoint and the real-time dynamic junction temperature feedback value. The controller's control parameters, including the proportional gain, integral time constant, and derivative time constant, are not fixed. Based on the real-time thermal resistance value output by the online thermal resistance network identification module in step 4, the system selects the most suitable set of control parameters from a preset gain scheduling table. This gain scheduling mechanism ensures that the stability margin of the closed-loop control system remains within the optimal range regardless of the aging condition of the test fixture or the environmental conditions.

[0056] When the dynamic junction temperature deviates from the preset target temperature control curve, the outer loop regulator generates a voltage correction signal that is superimposed on the input of the inner loop controller. Upon receiving the command, the fluid actuator drives the three-way proportional mixing valve. This actuator includes a high-power circulating pump to maintain constant flow circulation of the test medium within a vacuum-insulated pipeline; a heating unit with a high-power-density heating element; and a refrigeration unit based on compressor cooling.

[0057] The three-way proportional mixing valve precisely adjusts the cross-sectional area ratio of the low-temperature fluid to the high-temperature fluid entering the heat exchanger, achieving millisecond-level regulation of the output fluid temperature. The heat exchanger employs a highly thermally conductive microchannel structure, tightly fitting the test socket and transferring temperature changes to the chip surface via forced convection heat transfer. Throughout the testing process, the system continuously monitors airflow velocity, ambient humidity, and contact pressure. If the ambient humidity exceeds the dew point warning value, the system automatically increases the flow rate of the nitrogen purging device, forming a dry inert gas protective layer in the test area to prevent low-temperature condensation. Finally, through deep coupling of feedforward and dual-loop feedback, the fluctuation of the junction temperature inside the memory chip is strictly locked within ±0.5℃.

[0058] Example 2: In another preferred embodiment, the temperature control and monitoring method for the FT testing process of the memory chip described in this invention has been specifically optimized and extended for multi-station parallel testing scenarios. In this embodiment, the electrothermal dual-domain coupling feature model is designed as a parallel structure with multi-channel decoupling capability.

[0059] In step 1, the model considers the thermal characteristics of a single chip and also introduces a thermal interference matrix between adjacent test stations. By simulating the thermal conduction of the physical layout of the test interface units, the contribution rate of each station to the thermal radiation and conduction of adjacent stations under full load is quantified. This thermal coupling coefficient is integrated into the three-dimensional matrix feature table.

[0060] In the parameter acquisition stage of step 2, a multi-channel synchronous sampling architecture is adopted. The temperature control system has a built-in data distribution and processing core driven by a high-performance field-programmable gate array, which can simultaneously receive current and voltage waveform data from 32 or more test stations. Each channel is equipped with an independent digital-to-analog converter to ensure consistent sampling time.

[0061] During the electrothermal virtual mapping calculation in step 3, the processor employs parallel pipelined processing logic. For each chip, transient power consumption is calculated based on its independent operating state, and global temperature correction is performed in conjunction with the thermal interference matrix. This means that the dynamic junction temperature calculation for station A will reference the real-time power consumption states of stations B and C, eliminating the thermal superposition effect when multiple chips are tested simultaneously.

[0062] In the online thermal resistance network identification in step 4, the system introduces a collaborative filtering mechanism. If multiple adjacent physical sensors simultaneously detect similar temperature shift trends, the system algorithm automatically attributes it to changes in the ambient background temperature rather than a drift in the physical characteristics of a single test fixture. This logic reduces false compensation caused by environmental disturbances.

[0063] In step 5, the feedforward compensation instruction includes the linkage control logic for the cooling fans of the test equipment. The temperature control system calculates the total heat transfer intensity required for the surface of the test interface unit based on the total predicted power consumption of all workstations. The system dynamically adjusts the speed of the cooling fan array by outputting multiple pulse width modulation signals. When it is predicted that the overall test sequence is about to enter a sustained high-load phase, the fan speed is increased in advance to reduce the ambient reference thermal resistance.

[0064] In the closed-loop regulation of step 6, the fluid actuator employs a distributed flow distribution scheme. A miniature electronically controlled throttling valve is installed below each test station. The main circulation pump provides a constant fluid pressure, and the central controller independently adjusts the opening of each branch throttling valve based on feedback from the virtual junction temperature of each station. This architecture, combining global temperature-controlled medium preparation with precise local branch adjustment, enables independent and accurate temperature control for memory chips with different heat dissipation characteristics in large-scale mass production environments, enhancing the flexibility of the testing equipment when handling heterogeneous products.

[0065] Example 3: In an implementation scheme for testing ultra-high reliability memory chips, the method of the present invention adds a self-learning update mechanism. After the model is established as described in step 1, the system enters a continuous learning and evolution mode.

[0066] During the mass production tasks performed by the test equipment, the controller's internal logging module continuously collects junction temperature fluctuation curves, power consumption characteristic sequences, and historical trajectories of thermal resistance parameters identified online for each test cycle. This data is stored in a large-scale data buffer, and during idle periods such as equipment downtime or shift changes, the built-in lightweight machine learning algorithm is invoked.

[0067] This algorithm identifies structural biases in the model by comparing the dispersion between the observed temperature trajectory and the model's predicted trajectory. For example, if the system detects that a specific chip model consistently predicts a junction temperature slightly lower than the actual sensor reading when executing a particular compression write vector, the algorithm will automatically increase the thermal resistance weighting coefficient for that operating mode.

[0068] The self-learning mechanism also includes online modeling of the dynamic characteristics of the fluid actuator. The system periodically measures the response delay time from the issuance of the command to the actual change in fluid temperature. If it finds that the response delay has increased due to media aging or pipe scaling, the system will automatically increase the strength of the lead derivative compensation in the PID controller, or increase the number of pre-trigger clock cycles in the feedforward calculation in step 5.

[0069] To address the uneven heat generation within the memory chip, this embodiment introduces layout hotspot dynamic tracking technology in step 3. By interconnecting with the memory chip's design database, the system can obtain real-time information about the physical address space being activated by the current test vector. For read / write operations concentrated at the chip edge or in areas with poor thermal conductivity, the system automatically adjusts the spatial location coefficients in the virtual mapping algorithm to calculate a more representative hotspot junction temperature.

[0070] In step 6, an auxiliary thermoelectric cooler assembly is added to the actuator. This assembly is installed between the fluid heat exchanger and the test socket, utilizing the Peltier effect to provide extremely high bandwidth temperature fine-tuning. When a tiny, extremely high-frequency disturbance occurs in the dynamic junction temperature due to a sudden change in the test vector, the fluid control loop cannot respond due to mechanical inertia. At this time, the thermoelectric cooler receives a high-frequency control voltage and, by changing the direction and magnitude of the current, instantly absorbs or releases heat, thus suppressing millisecond-level junction temperature fluctuations. This combination of wide-range fluid regulation and precise thermoelectric fine-tuning further improves the temperature control accuracy to the limit level of ±0.1℃.

[0071] The methods described above also involve continuous safety monitoring of the testing environment. The system is equipped with a multi-physical quantity sensing matrix to monitor the airflow distribution, ambient humidity percentage, and contact pressure sensor readings at the four corners of the test socket in real time. When an imbalance is detected in the four signals output by the contact pressure sensors, it is determined that the chip placement is off-center or the test socket pressure head is tilted. The system immediately interrupts the feedforward control command and issues a mechanical structure warning to prevent chip burnout due to abnormal heat conduction.

[0072] During the low-temperature testing phase, the nitrogen purging device's logic is fully integrated with the junction temperature control. The system calculates the dew point corresponding to the current heat exchanger surface temperature and dynamically adjusts the nitrogen flow rate based on real-time humidity data to ensure the test area remains dry at all times. When the test task is completed and the chip needs to be ejected from the test socket, the system automatically activates the rapid temperature recovery logic, driving the fluid actuator to raise the temperature to room temperature within seconds, preventing condensation of ambient moisture due to overcooling during chip removal.

[0073] The electrothermal dual-domain coupled characteristic model maintains periodic self-calibration throughout the equipment's lifecycle. Whenever a critical mechanical component or temperature control medium is replaced, the system forces a full-range thermal characteristic scan. By executing standard power excitation sequences under no-load and full-load conditions, the system automatically regenerates the multi-dimensional feature library, ensuring continuous consistency in control performance.

[0074] Example 4: In an application scenario for extreme temperature cycling testing of automotive-grade memory chips, the method described in this invention has been optimized for wide temperature range switching performance.

[0075] In step 1, the thermal property calibration covers the entire temperature range from -55 degrees Celsius to +150 degrees Celsius. Because the thermal diffusivity of the encapsulation material varies drastically at different temperatures, the three-dimensional matrix feature table is divided into multiple temperature-specific sub-tables. The system automatically switches to the corresponding sub-table for calculation based on the current ambient reference temperature.

[0076] In step 2, considering the temperature drift of electronic components under extremely high temperature environments, the current monitoring circuit adds an automatic zero-point correction function. During the microsecond interval before the start of each test sequence, the system samples the reference current under no-load conditions and subtracts this bias from the subsequent real-time sampled values.

[0077] In step 3, the virtual mapping algorithm adds a thermal radiation loss compensation term. During the high-temperature testing phase, the temperature difference between the chip and the surrounding environment is huge, and the heat dissipated through radiation is not negligible. The processor calculates the contribution of radiative heat transfer from the chip surface to the surrounding space based on a textual variant of the Stefan-Boltzmann law, and subtracts it from the temperature rise contribution caused by total power consumption to correct the estimated junction temperature.

[0078] In step 4, the thermal resistance network identification module adds online estimation of pipeline cold loss. During low-temperature testing, the path between the heat exchanger and the chip absorbs heat from the external environment, leading to a decrease in cooling efficiency. The Kalman filter identifies the performance status of the pipeline insulation layer by monitoring the time lag between the fluid outlet temperature and the chip casing temperature.

[0079] In the feedforward logic of step 5, the system introduces multi-level power consumption ladder prediction. For complex stress test vectors, the system not only predicts the power consumption at the next jump point, but also anticipates the power consumption envelope over the next 100 clock cycles. Based on this envelope, the fluid control command is optimized into a smooth flow rate change curve, avoiding mechanical wear and fluid oscillations caused by frequent start-stop of the proportional valve.

[0080] In step 6, the rotational speed of the circulating pump is set to a variable positively correlated with the real-time thermal resistance value. When an increase in thermal resistance is detected, the pump speed is automatically increased to enhance turbulence intensity and improve the convective heat transfer coefficient. Simultaneously, for extremely low-temperature environments, the heating unit is switched to anti-freeze protection mode to ensure that the circulating medium does not solidify.

[0081] The electrothermal virtual mapping algorithm and thermal resistance identification logic involved in the above embodiments are all executed by high-priority tasks in the embedded real-time operating system to ensure the deterministic response of the control loop. All data interactions are conducted through a serial bus with error detection and correction functions to ensure that temperature control and monitoring during the FT test of the memory chip are always in a high-safety and high-precision operating state.

[0082] Through the implementation of the above embodiments, this invention transforms the temperature control system from a traditional passive response to ambient temperature to an active sensing and prediction of changes in the chip's internal junction temperature. By deeply fusing electrical and thermal signals, it solves the long-standing problem of heat transfer hysteresis in memory chip FT testing. Utilizing the chip's own electrical parameters as a virtual sensor not only reduces the complexity of the hardware structure but also ensures robustness and consistency in the testing process under complex industrial production environments through online identification and self-learning mechanisms, ultimately improving the production line yield and testing efficiency of memory chips.

[0083] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for temperature control and monitoring during the FT test process of a memory chip, characterized in that, Includes the following steps: Step 1: Establish an electrothermal dual-domain coupling feature model. By calibrating the thermal characteristics of the memory chip, obtain and record the mapping relationship between the power consumption, structural thermal resistance and internal junction temperature of the memory chip in different working modes, and form a multi-dimensional feature library. Step 2: Real-time acquisition of electrical operating parameters. During the testing of the memory chip, the transient operating current and transient operating voltage of the memory chip during the test vector operation are detected and recorded by the high-speed synchronous sampling unit. Step 3: Perform electrothermal virtual mapping calculation, calculate the real-time power consumption using the collected transient operating current and transient operating voltage, retrieve the mapping relationship using the preset calculation logic, and solve the dynamic junction temperature inside the memory chip in real time using the virtual mapping algorithm; Step 4: Implement online thermal resistance network identification, compare the calculated dynamic junction temperature with the chip casing temperature collected by the physical sensor, and use an adaptive filtering algorithm to dynamically identify the real-time thermal resistance network parameters between the chip junction and the test environment. Step 5: Generate feedforward compensation control instructions, receive test item jump signals issued by the test machine in real time, obtain the time node of power consumption change in advance, and calculate the predicted junction temperature change based on the electrothermal virtual mapping to generate compensatory fluid flow adjustment instructions. Step 6: Perform closed-loop feedback regulation, using the dynamic junction temperature as the main feedback quantity and the chip casing temperature as the auxiliary feedback quantity. A control signal is generated by the regulator to drive the fluid actuator to adjust the mixing ratio of the medium, thereby locking the junction temperature inside the memory chip.

2. The temperature control and monitoring method for the FT test process of a memory chip according to claim 1, characterized in that, The process of establishing the electrothermal dual-domain coupling feature model in step 1 includes: using thermal simulation software to establish a fine thermal model of the memory chip, and performing mesh processing on the silicon wafer, lead frame, molding resin, solder ball array, and supporting test socket and test carrier board inside the memory chip. Extract the thermal conductivity, specific heat capacity, coefficient of thermal expansion, and density parameters corresponding to each packaging material from the material property library, and set the functional relationship between the parameters and temperature. The thermal diffusion path of the memory chip under different packaging materials and pin arrangements is analyzed, the physical positions of each power pin and ground pin are extracted, the parasitic resistance heating of each pin during signal transmission is calculated, and it is loaded into the fine thermal model as a local heat source. The contact conditions between the test base and the surface of the memory chip under different pressure gradients were simulated to establish a contact thermal resistance model and obtain the negative correlation mapping curve between contact thermal resistance and mechanical pressure. In the experimental environment, a step-by-step power load was applied to the sample chip, and the chip shell temperature and the readings of the internal temperature sensor were recorded at different ambient temperature points to obtain the nonlinear compensation coefficient of thermal resistance as a function of temperature. The simulation data of the refined thermal model is fused with the experimental calibration data to form a feature table with operating voltage, operating current, and ambient temperature as input variables and junction temperature as output variable, and the feature table is stored in a non-volatile memory unit.

3. The temperature control and monitoring method for the FT test process of a memory chip according to claim 2, characterized in that, The process of acquiring electrical operating parameters in real time in step 2 includes: connecting a current monitoring circuit in series in the power supply channel of the memory chip; the current monitoring circuit uses a Hall effect sensor or a precision sampling resistor in conjunction with an instrumentation amplifier to acquire the current waveform of the power supply pin of the memory chip. The input voltage of the memory chip is monitored by a voltage comparator to obtain the voltage ripple change; The sampling frequency of the high-speed synchronous sampling unit is synchronized with the clock cycle of the test instrument to ensure that the electrical characteristics corresponding to each test vector of the memory chip are completely captured. The sampling control logic is synchronized with the trigger signal of the test instrument at the hardware level. Whenever the test instrument sends a start trigger pulse, the high-speed synchronous sampling unit starts the sampling mode. The sampled transient operating current and transient operating voltage are cached in random access memory and then subjected to first-order hysteresis filtering before entering the calculation logic to filter out high-frequency pulse noise.

4. The temperature control and monitoring method for the FT test process of a memory chip according to claim 3, characterized in that, The process of performing the electrothermal virtual mapping calculation in step 3 includes: multiplying the real-time collected transient operating current and the transient operating voltage to obtain the total transient power consumption of the memory chip in the current test cycle; The task status word from the test machine is parsed to identify the current logical operation mode of the memory chip, which includes standby mode, continuous read operation mode, page programming operation mode or whole chip erase mode. Based on the identified logical operation mode, the corresponding thermal resistance reference value is retrieved from the multi-dimensional feature library; The product of the transient total power consumption and the thermal resistance reference value is used as the temperature increment. The temperature increment is then superimposed on the reference temperature measured by the physical sensor to obtain the initial estimate of the dynamic junction temperature. The initial estimate of the dynamic junction temperature is smoothed in the time domain using a thermal capacity correction factor to compensate for the phase shift caused by the thermal inertia of the packaging material. The phase shift compensation process is achieved by calculating the difference between the initial estimate at the current moment and the dynamic junction temperature at the previous moment, and multiplying it by an attenuation coefficient related to the thermal time constant of the packaging.

5. The temperature control and monitoring method for the FT test process of a memory chip according to claim 4, characterized in that, The virtual mapping algorithm in step 3 further includes: compensation logic for uneven heating in multiple parts inside the memory chip, which divides the chip into a memory array area, a logic control area and a peripheral interface area according to the internal layout design of the memory chip; Different weighting coefficients are assigned based on the activation levels of different functional areas by different test vectors; When performing parallel read tests, increase the weighting coefficient of the storage array area; when performing serial interface calibration tests, increase the weighting coefficient of the peripheral interface area. The local temperature distribution of each functional area is calculated, and the equivalent junction temperature value is obtained by weighted average algorithm.

6. The temperature control and monitoring method for the FT test process of a memory chip according to claim 5, characterized in that, The process of implementing online thermal resistance network identification in step 4 includes: constructing a thermal resistance observer based on state-space equations, using the dynamic junction temperature as a state variable and the chip casing temperature as an observation variable; The Kalman filter algorithm is used to perform prediction and update steps within the sampling period; In the prediction step, the predicted value of thermal resistance and the predicted value of covariance matrix at the current moment are calculated using the thermal resistance state value and the heat transfer state transition matrix at the previous moment. In the update step, the physical sensor is used to collect the measured value of the chip casing temperature, and the observation residual between the measured value and the predicted casing temperature is calculated. The gain coefficient is calculated based on the state prediction covariance and the observation noise covariance, and the observation residual is weighted using the gain coefficient to correct the thermal resistance prediction value. The identification process identifies thermal resistance drift caused by aging of the metal probe of the test seat, fluctuation of contact pressure, or changes in humidity of the test environment, and feeds back the corrected thermal resistance network parameters to the calculation process in step 3.

7. The temperature control and monitoring method for the FT test process of a memory chip according to claim 6, characterized in that, The process of generating feedforward compensation control instructions in step 5 includes: the test equipment sends a pre-trigger pulse signal to the temperature control system during a preset clock cycle before the high-power test item jumps; After the temperature control system analyzes the pre-trigger pulse signal, it extracts the expected power consumption level for the next stage from the test vector sequence. Calculate the difference between the expected power consumption level and the current power consumption level, and combine the electrothermal virtual mapping to calculate the junction temperature fluctuation amplitude caused by the difference; Using the junction temperature fluctuation amplitude caused by the difference and combined with the inversion algorithm of the heat conduction model, the compensation flow rate required by the fluid actuator is derived; The opening of the proportional valve is adjusted in advance to counteract the transient changes in the heat generated by the memory chip through proactive control.

8. The temperature control and monitoring method for the FT test process of a memory chip according to claim 7, characterized in that, The process of performing closed-loop feedback regulation in step 6 includes: constructing a dual-loop control architecture, wherein the inner loop is a fluid flow control loop and the outer loop is a temperature compensation control loop; The proportional gain, integral time constant, and derivative time constant of the regulator are dynamically adjusted based on the real-time thermal resistance network parameters identified and output online by the thermal resistance network. When the dynamic junction temperature deviates from the preset target value, the regulator of the outer loop generates a correction signal that is superimposed on the input terminal of the regulator of the inner loop. The mixing ratio of cryogenic fluid and high-temperature fluid is adjusted by the fluid actuator to ensure that the internal junction temperature of the memory chip remains within a predetermined threshold range throughout the entire testing process.

9. The temperature control and monitoring method for the FT test process of a memory chip according to claim 8, characterized in that, The fluid actuator includes a circulating pump, a heating unit, a cooling unit, a three-way proportional mixing valve, a microchannel heat exchanger, and a cooling fan; The circulating pump maintains the test medium in a constant flow circulation in the closed pipeline, and the heating unit and the cooling unit respectively maintain the test medium at a preset high temperature state and a low temperature state. The three-way proportional mixing valve allocates the ratio of hot and cold media entering the microchannel heat exchanger according to the control signal. The microchannel heat exchanger contacts the test socket and heats or cools the memory chip through heat conduction. When it is predicted that the memory chip will enter a high-power mode, the pulse width modulation duty cycle of the cooling fan is increased simultaneously to enhance the convective heat transfer intensity on the surface of the test interface unit.

10. The temperature control and monitoring method for the FT test process of a memory chip according to claim 9, characterized in that, It also includes: using a multi-physical quantity sensing matrix to continuously monitor the airflow velocity, ambient humidity and test seat contact pressure inside the test chamber; When the ambient humidity is detected to exceed the predetermined range, the nitrogen purging device is activated, and a dry inert gas protective layer is formed in the test area. When the test seat contact pressure is detected to be lower than a predetermined threshold or there is a pressure imbalance, an early warning signal is issued and the feedforward compensation control command is interrupted. Before the test task ends and the chip is ejected, the temperature recovery logic is activated to drive the fluid actuator to raise the temperature to room temperature. The electrothermal dual-domain coupling feature model is periodically updated through self-learning. By collecting junction temperature fluctuation data and electrical parameter features, machine learning algorithms are used to identify model deviations and correct the parameters in the multi-dimensional feature library.

Images