Semiconductor high-temperature gate bias test system and method based on dynamic temperature control and intelligent elimination

Abstract

The invention discloses a semiconductor high-temperature gate bias test system and method based on dynamic temperature control and intelligent elimination. The system comprises a partition temperature control test box, a multi-path progressive programmable power supply, a high-side switch station array, a high-speed data acquisition unit and an intelligent control center. According to the method, the uniformity and stability of a space temperature field in a test box are ensured through a dynamic partition temperature control algorithm; through a voltage gradual output and overshoot suppression strategy, damage of test voltage sudden change to a device is avoided; through independent station control and millisecond-level real-time data acquisition based on a high-side switch and in combination with a current trend prediction model based on a long short-term memory (LSTM) network, intelligent early warning and rapid and accurate elimination of abnormal devices are realized. The problems of uneven temperature, voltage overshoot, station linkage interference, slow abnormal response and the like in the prior art are solved, and the reliability, the accuracy and the automation degree of the high-temperature gate offset test are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device reliability testing technology, and in particular to a system and method for performing high-temperature gate bias (HTGB) tests on discrete semiconductor devices such as diodes, transistors, MOSFETs, and IGBTs. Background Technology

[0002] High-temperature gate bias (HTGB) testing is a crucial method for evaluating the long-term reliability of semiconductor devices under high temperature and high voltage stress. The accuracy of the test results directly affects the quality assessment and lifespan prediction of the devices. Existing HTGB test systems typically consist of a high-temperature test chamber, a DC bias power supply, an aging board with loaded devices, and a data acquisition system.

[0003] However, existing technologies generally suffer from the following drawbacks: Poor temperature uniformity control: Traditional test chambers often use a single heating source in conjunction with a fan for overall temperature field control. Under full load conditions, this method results in significant temperature gradients (potentially exceeding ±5℃) at different locations within the test chamber due to uneven heat convection and differences in component heating. This temperature inconsistency introduces additional test variables, making the failure rates of components at different locations incomparable, severely impacting the accuracy and reliability of test data.

[0004] Test power supply voltage overshoot: If the programmable power supply outputs in a step manner at the instant the test voltage is applied, it is very easy to generate voltage overshoot due to parasitic parameters such as line inductance and capacitance. This instantaneous high voltage spike will cause irreversible damage to the fragile gate oxide layer or PN junction, causing the device to fail in the early stage of the test, thus failing to truly reflect its life characteristics under actual operating conditions.

[0005] The current station control is crude and prone to mutual interference: Most existing systems use a group power supply mode, where one power channel simultaneously supplies power to multiple or even dozens of stations. When a device in one station experiences an anomaly such as breakdown or increased leakage current, the abnormal current will lower the voltage of the entire power supply group, or the shared current-limiting resistor may overheat, affecting the test conditions of other normal devices in the same group, resulting in a "one failure, all failures" situation. Although some systems have introduced rejection functions, the response speed is slow (usually on the order of seconds or more), and they are mostly simple threshold comparisons, unable to predict progressive failures.

[0006] Lagging data acquisition and processing capabilities: For large-scale parallel experiments with thousands of workstations, traditional data acquisition systems suffer from low sampling rates and limited channel counts, making it difficult to achieve near real-time monitoring of all workstations. Data analysis is typically limited to post-event threshold judgments, lacking intelligent analysis and early warning capabilities based on data trends.

[0007] Therefore, there is an urgent need in this field for a high-temperature grid bias test system and method that can achieve high-precision uniform temperature control, no overshoot voltage excitation, completely independent control and monitoring of the workstation, and intelligent early warning capability. Summary of the Invention

[0008] Purpose of the invention: The primary purpose of this invention is to overcome the shortcomings of the prior art and provide a semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection.

[0009] Another object of the present invention is to provide a test method applicable to the above-described system.

[0010] Technical solution: A high-temperature gate bias testing system for semiconductors based on dynamic temperature control and intelligent rejection, including: The zoned temperature control test chamber is divided into multiple independent temperature zones. Each temperature zone is equipped with an independent heater and temperature sensor, all connected to a temperature control processor. A multi-channel progressive programmable power supply is used to provide the DC bias voltage required for the test, and its output voltage follows a preset progressive rise curve. The high-side switch station array consists of several aging boards. Each aging board contains multiple independent test stations, and each station is equipped with a high-side switch module and a current sampling module. A high-speed data acquisition unit is connected to the current sampling module and the temperature sensor to synchronously acquire current data from all workstations and temperature data from each temperature zone. The intelligent control center runs test control software, which integrates a dynamic zone temperature control module, a power control module, a workstation on / off control module, and an intelligent rejection decision module.

[0011] In a further embodiment, the dynamic zone temperature control module executes a zone temperature control algorithm. This algorithm calculates and adjusts the output power of the heater in each zone independently based on the difference between the temperature value fed back by the temperature sensor in each temperature zone and the target temperature through an adaptive PID controller.

[0012] In a further embodiment, the multi-channel progressive programmable power supply adopts an S-curve voltage ramp control strategy, and its output voltage function is:

[0013] in To set the voltage, This is a time constant that is dynamically configured based on the characteristics of the device under test.

[0014] In a further embodiment, the intelligent rejection decision module includes an LSTM-based current trend prediction unit. This unit takes the real-time and historical current data sequence of the workstation as input, predicts the current value within a specific future time window, and if the predicted value exceeds a preset threshold, sends a rejection command to the high-side switch module of the workstation.

[0015] In a further embodiment, the rejection response time of the intelligent rejection decision module is no more than 50 milliseconds.

[0016] A semiconductor high-temperature gate bias test method based on dynamic temperature control and intelligent rejection, applied to the system described above, characterized in that the method includes: Dynamic zone temperature control steps; Progressive voltage output steps; Steps for independent on / off control and status monitoring of workstations; Intelligent early warning and elimination steps based on LSTM.

[0017] In a further embodiment, the LSTM-based intelligent early warning and elimination step includes: Construct and train an LSTM neural network model, whose input is the time series data of the station current and whose output is the predicted value of the current in the future period; During the experiment, the current data of each station was input into the trained LSTM model in real time. The model's current predictions are compared with the preset alarm current threshold and rejection current threshold. When the predicted value exceeds the alarm current threshold, an early warning record is made; when the predicted value exceeds the rejection current threshold, a rejection signal is immediately generated and the bias voltage supply to that station is cut off.

[0018] Compared with the prior art, the present invention has the following significant advantages: 1. Extremely high accuracy of test data: Through dynamic zone temperature control, it is ensured that all devices in all locations within the test chamber are under highly consistent thermal stress, eliminating test errors introduced by uneven temperature, and making failure data more comparable and reliable.

[0019] 2. Superior device protection capability: The progressive voltage output fundamentally eliminates damage to devices caused by voltage overshoot, making it particularly suitable for advanced process devices that are sensitive to voltage spikes.

[0020] 3. True workstation independence and high reliability: The high-side switch design makes each workstation an independent test unit. Abnormal workstations are physically isolated, completely eliminating mutual interference between workstations and greatly improving the overall reliability of the system.

[0021] 4. Intelligent and Foresightful: The introduction of an LSTM model for current trend prediction enables a shift from "post-event remediation" to "pre-event warning." It can identify failure trends and intervene early before the device current reaches a dangerous threshold. This not only protects the test system but, more importantly, captures valuable data on early device failures, which is crucial for failure analysis.

[0022] 5. High efficiency and low maintenance cost: The fast rejection response (50ms) and accurate early warning avoid the chain reaction that may be caused by the complete failure of a single component (such as burning out the PCB, carbonizing the current limiting resistor, etc.), reducing the system maintenance frequency and consumable costs. Attached Figure Description

[0023] Figure 1 This is a diagram of the overall hardware architecture of the system of this invention.

[0024] Figure 2 This is a flowchart of the dynamic zone temperature control algorithm of the present invention.

[0025] Figure 3 This is a graph showing the progressive voltage output of the present invention.

[0026] Figure 4 This is a flowchart of the intelligent early warning and elimination decision-making process based on LSTM of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] A semiconductor high-temperature gate bias testing system and method based on dynamic temperature control and intelligent rejection, such as Figure 1 As shown, the system provided by the present invention includes: 1. Zoned Temperature Control Test Chamber The interior of the test chamber is physically or logically divided into N (e.g., 16, corresponding to 16 slot areas) independent temperature zones. Each temperature zone is configured with: High-precision platinum resistance temperature sensor (PT100): Real-time monitoring of the actual temperature of the area.

[0029] Independently controlled heaters: such as armored heating rods, whose power is independently adjusted by a temperature control processor.

[0030] Circulating air duct design: By optimizing the air duct, the uniform flow of air inside the chamber is promoted, and static temperature dead zones are reduced.

[0031] All temperature zone sensors and heaters are connected to a central temperature control processor, which runs a dynamic zone temperature control algorithm.

[0032] 2. Multi-channel progressive programmable power supply The system provides multiple (e.g., 3) independently programmable high-voltage DC power supplies with an output voltage range of 0-2000V. Its core innovation lies in the adoption of a progressive output strategy and overshoot suppression circuitry. Under software control, the power supply output voltage does not instantly reach the set value, but rather slowly rises along a smooth S-shaped curve (or its approximate curve), effectively suppressing inrush current and voltage overshoot.

[0033] 3. High-side switch station array This is the core of achieving independent control of each workstation. The system contains multiple aging board slots, and each aging board has M (e.g., 80) completely independent testing workstations. Each workstation includes: High-side switching module: Typically consists of a high-voltage MOSFET and its driving circuit, connected in series between the positive terminal of the power supply and the positive terminal of the device. By controlling the gate of this MOSFET, the bias voltage of this station can be independently turned on or off.

[0034] Current sampling module: Employs a high-side current detection amplifier to accurately measure the minute current (as low as 0.01nA) flowing through the device at this workstation.

[0035] This "high-side control" architecture ensures that when a station is removed due to an anomaly, its load is completely removed from the test loop without affecting the test conditions of other stations.

[0036] 4. High-speed data acquisition unit This unit consists of multiple high-precision data acquisition cards, featuring high resolution (e.g., 24-bit) and high sampling rate (the overall system sampling rate can reach 1MHz). It simultaneously acquires current data from all 1280 workstations, voltage data from each power supply, and temperature data from each temperature zone, providing a data foundation for real-time monitoring and intelligent decision-making.

[0037] 5. Intelligent Control Center Typically, this is an industrial computer running dedicated test control software. In addition to providing a user interface, this software integrates the following core intelligent modules: Dynamic zone temperature control module: executes zone temperature control algorithm.

[0038] Power control module: manages the output of progressive voltage.

[0039] Workstation on / off control module: Provides an interface for batch setting of workstation on / off states (such as full on, every other workstation on, etc.).

[0040] Intelligent rejection decision module: This is the intelligent embodiment of the invention, and its core is a current trend prediction model based on LSTM.

[0041] like Figure 2 and Figure 3 As shown, the main steps of the method of the present invention include: S1: System Initialization and Parameter Settings Users can set parameters such as target temperature (e.g., 175℃), test voltage (e.g., 1500V), test duration, and alarm / rejection current threshold through the software interface. The system self-checks the status of each module.

[0042] S2: Dynamic zone temperature control Temperature control processor reads all Real-time temperature values ​​for each temperature zone .

[0043] For each temperature zone, calculate its temperature deviation. .

[0044] The adaptive PID control algorithm is invoked to calculate the output power of the heater in this temperature range. :

[0045] in, , , The parameters can be tuned online according to the thermal characteristics of the temperature range to optimize the dynamic response and avoid overshoot.

[0046] The system continuously operates this closed-loop control to ensure that the temperature in all temperature zones matches the target temperature throughout the entire test. The deviation is stable within ±1℃.

[0047] S3: Progressive voltage output After the user starts the power supply output, the power control module does not directly output the target voltage. Instead, it generates real-time voltage commands according to the following function. :

[0048] Wherein, time constant It is an adjustable parameter; for sensitive devices, a larger setting can be used. A value (such as 1 second) makes the voltage rise more gradually.

[0049] The internal feedback control loop of the power supply accurately tracks This achieves a smooth, non-overshoot-induced rise in output voltage.

[0050] S4: Independent on / off and status monitoring of workstations Before applying voltage, users can use software to perform batch operations to put the high-side switches of all stations or designated stations into the "conducting" state.

[0051] The high-speed data acquisition unit begins to collect current data from all workstations at a high rate. The results are displayed in real time on the software interface.

[0052] S5: LSTM-based intelligent early warning and elimination This step is the core innovation of the method of this invention, such as... Figure 4 As shown, the detailed process is as follows: Model preparation: An LSTM neural network model was pre-trained using historical HTGB test data (including current time series for normal and various failure modes). This model is able to learn normal and abnormal current patterns over time.

[0053] Real-time prediction: During the experiment, for each workstation, the intelligent elimination decision module continuously acquires the current data of its most recent L sampling points, forming a time series window. .

[0054] The sequence is input into a trained LSTM model, which outputs the predicted current values ​​for the next K time points. .

[0055] Decision-making and execution: Warning: If any of the predicted values If the current exceeds the user-defined alarm current threshold (e.g., 200nA) but has not yet reached the rejection current threshold, the system will mark the workstation as "warning" in the software interface and record it in the log to remind the operator to pay attention, but will not immediately cut off the voltage.

[0056] Elimination: If any predicted value If the current exceeds the user-defined rejection threshold (e.g., 2000nA), the system immediately determines that the device at that station is about to experience or has already experienced a fatal failure. The intelligent rejection decision module sends a shutdown signal to the high-side switch module at that station within a very short time (≤50ms), physically cutting off its bias supply. This process is independent and rapid, and does not affect any other stations.

[0057] Continuous learning: The system can feed newly generated abnormal data back to the model for online updates and optimization, enabling self-evolution.

[0058] The present invention will be further described below with reference to a specific, non-limiting embodiment.

[0059] Example: A 1000-hour HTGB test was conducted on a certain type of IGBT. Preparation: The operator mounts 80 IGBT devices of this model onto an aging board and inserts the aging board into slot 1 of the test chamber. A new test program is created in the software's "Device Library," and the parameters are set as follows: test temperature. Test voltage Test duration = 1000 hours, alarm current = 100nA, rejection current = 1000nA.

[0060] Temperature Control Startup: The user initiates the test. The dynamic zone temperature control module begins operation. The temperature control processor reads the temperature of the zone containing slot 1 and all other zones, and controls the power of each heater using an adaptive PID algorithm. After approximately 30 minutes, the system reaches thermal stability, with the temperature at each point stabilizing at 150±0.8℃.

[0061] Apply bias voltage: The user sets all 80 stations in slot 1 to "On" via the "Station On / Off" interface. Then, power channel 1 is activated via the "Test Power Supply" interface. The power supply output voltage follows... The curve smoothly rises to 1200V in about 4 seconds, and no observable overshoot is observed when measured with an oscilloscope.

[0062] Intelligent Monitoring and Rejection: At 500 hours of the experiment, the high-speed data acquisition system continuously monitored the current at all workstations. At this point, the intelligent rejection decision module detected a slight upward trend in the current sequence of workstation number 35. Based on current data from the previous few hours, the LSTM model predicted that the current at this workstation might exceed 1000nA within the next hour.

[0063] When the actual current value is only 150nA (far below the rejection threshold), the system issues an early warning and records it on the interface.

[0064] Approximately 40 minutes later, the current at this workstation rose to 800nA, and the LSTM model updated its prediction, confirming that it would exceed 1000nA within 10 minutes.

[0065] The system immediately activated, cutting off the high-side switch of station 35 within 50ms. At this point, the current at that station dropped to zero, while the current and voltage at the other 79 stations remained unaffected, and the test continued smoothly.

[0066] Data and Reports: After the test, users can use the "Data / Curve" function to view the historical current curves of all workstations and export detailed reports containing test parameters, temperature and humidity records, and abnormal events.

[0067] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection, characterized in that, include: The zoned temperature control test chamber is divided into multiple independent temperature zones. Each temperature zone is equipped with an independent heater and temperature sensor, all connected to a temperature control processor. A multi-channel progressive programmable power supply is used to provide the DC bias voltage required for the test, and its output voltage follows a preset progressive rise curve. The high-side switch station array consists of several aging boards. Each aging board contains multiple independent test stations, and each station is equipped with a high-side switch module and a current sampling module. A high-speed data acquisition unit is connected to the current sampling module and the temperature sensor to synchronously acquire current data from all workstations and temperature data from each temperature zone. The intelligent control center runs test control software, which integrates a dynamic zone temperature control module, a power control module, a workstation on / off control module, and an intelligent rejection decision module.

2. The semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection according to claim 1, characterized in that, The dynamic zone temperature control module executes a zone temperature control algorithm. This algorithm calculates and adjusts the output power of the heater in each zone independently based on the difference between the temperature value fed back by the temperature sensor in each zone and the target temperature through an adaptive PID controller.

3. The semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection according to claim 1, characterized in that, The multi-channel progressive programmable power supply adopts an S-curve voltage ramp control strategy, and its output voltage function is: ； in To set the voltage, This is a time constant that is dynamically configured based on the characteristics of the device under test.

4. The semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection according to claim 1, characterized in that, The intelligent rejection decision module includes an LSTM-based current trend prediction unit. This unit takes the real-time and historical current data sequence of the workstation as input and predicts the current value within a specific time window in the future. If the predicted value exceeds a preset threshold, a rejection command is sent to the high-side switch module of the workstation.

5. The semiconductor high-temperature gate bias test system based on dynamic temperature control and intelligent rejection according to claim 4, characterized in that, The intelligent rejection decision module has a rejection response time of no more than 50 milliseconds.

6. A semiconductor high-temperature gate bias test method based on dynamic temperature control and intelligent rejection, applied to the system described in any one of claims 1-5, characterized in that, The method includes: Dynamic zone temperature control steps; Progressive voltage output steps; Steps for independent on / off control and status monitoring of workstations; Intelligent early warning and elimination steps based on LSTM.

7. The semiconductor high-temperature gate bias test method based on dynamic temperature control and intelligent rejection according to claim 6, characterized in that, The LSTM-based intelligent early warning and elimination steps include: Construct and train an LSTM neural network model, whose input is the time series data of the station current and whose output is the predicted value of the current in the future period; During the experiment, the current data of each station was input into the trained LSTM model in real time. The model's current predictions are compared with the preset alarm current threshold and rejection current threshold. When the predicted value exceeds the alarm current threshold, an early warning is recorded; when the predicted value exceeds the rejection current threshold, a rejection signal is immediately generated and the bias voltage supply to that station is cut off.

Images