An AI chip single particle effect evaluation method

Abstract

The application provides an AI chip single particle effect evaluation method, relates to the technical field of integrated circuits, and comprises ten steps of a linear energy transfer threshold value for estimating single particle effect of an AI chip, preliminary test path selection, laser test path development, laser test data processing, secondary test path determination, proton test path development, proton test result determination, proton test data processing, heavy ion test parameter selection, and heavy ion test and result confirmation; the application uses a 10 MeV·cm²·mg ‑1 linear energy transfer threshold value as a demarcation for a combined evaluation system of pulsed laser, proton and heavy ion, adopts threshold value alignment or equal-probability equivalent mapping to realize cross-method result alignment, and forms a reproducible evidence closed loop by cooperating with minimum heavy ion point calibration, so that the comparability of conclusions is ensured, and machine time, cost and evaluation period are significantly reduced.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit technology, and in particular to a method for evaluating single-event effects in AI chips. Background Technology

[0002] In the space radiation environment, charged particles ionize along their tracks when passing through the region near the PN junction of a semiconductor device, generating a large number of charge carriers. These charges are rapidly collected under the influence of the device's built-in electric field. When the collected amount exceeds a critical charge, it may trigger an unexpected flip of the circuit's logic state or form a continuous high-current path, i.e., single-event upset (SEU) and single-event latch-up (SEL), and even induce single-event functional failure (SEFI), thus threatening the spacecraft's circuit system. To establish ground-based evidence of on-orbit reliability, engineering practice typically uses pulsed laser devices and proton and heavy-ion accelerators to characterize and evaluate the single-event effect (SEE) resistance of circuits or devices. In the future, more advanced process devices will be enabled to meet the needs of space applications. For example, the increasing computing power requirements of space-based intelligent processing equipment for new artificial intelligence (AI) applications in space will make the deployment of commercial AI devices an inevitable choice. AI chips face the following new challenges in resisting single-event effects: advanced nanoscale devices exhibit non-traditional radiation damage mechanisms, such as single-event pinpoint hard damage and single-event micro-latch-up; the influence of operating dynamic frequency on the characteristics of single-event effects is complex. Therefore, there are differences in the perspectives used to assess device reliability, and traditional reliability assessment standards and methods are not directly applicable to commercial aerospace AI chips.

[0003] Pulsed laser single-event testing technology has advantages such as readily available equipment, flexible configuration, precise positioning, and rapid scanning. Proton irradiation assessment also has advantages such as stable beam, high flux, tunable energy spectrum, and greater representativeness of the shielded environment. However, there are differences between lasers and real particles in terms of deposition volume and carrier distribution. The equivalence and reproducibility between lasers and heavy ion tests are easily affected by sample structure and process details. Proton irradiation can form real physical tracks and has stronger environmental representativeness and high flux advantages in the low linear energy transfer (LET) range. Unlike lasers, it does not require open packaging and its wide beam irradiation can cover the entire device surface. However, it cannot cover and achieve high energy density induced single-event effects, which can easily lead to incomplete conclusions and misjudgment risks when used alone. Although heavy ion is one of the most recognized calibration benchmarks, facilities are scarce, scheduling is tight, and costs are high. Moreover, the spatial resolution is limited under wide beam conditions, making it difficult to complete high-density layout-level screening on complex systems.

[0004] In traditional ground-based experiments, the aforementioned technical approaches are all conducted independently, without proposing a standardized joint evaluation process that uses a clear linear energy transmission threshold as a boundary, with lasers responsible for high-range baseline localization, protons responsible for statistics within the low-range threshold, and a small number of heavy ions for alignment. This leads to problems such as low efficiency, high machine time consumption, and difficulty in forming closed-loop conclusions in current engineering practices. Therefore, this invention proposes a single-event effect evaluation method for AI chips to solve the problems existing in the prior art. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to propose a single-event effect evaluation method for AI chips. This method uses the linear energy transfer (LET) threshold as a boundary, with pulsed lasers responsible for the baseline and localization of the high LET range, and protons responsible for the threshold determination and single-event effect incidence assessment of the low LET range. Furthermore, it achieves a joint evaluation method for single-event effects that aligns with heavy ion experimental results through equivalent calibration.

[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a method for evaluating single-event effects in AI chips, comprising the following steps:

[0007] Step 1: Estimate the linear energy transfer threshold for single-event effects in AI chips.

[0008] Based on the layout and process data of the AI ​​chip, typical workload, existing similar sample data or related tests, and combined with the charge collection model, the linear energy transfer threshold required for the AI ​​chip under test to produce a single-event effect is determined, and an estimated LET threshold and anomaly judgment rules are formed.

[0009] Step 2: Preliminary test path selection

[0010] The system sets a preset LET threshold boundary value, and then compares the estimated LET threshold with the preset LET threshold boundary value. When the estimated LET threshold is greater than or equal to the preset LET threshold boundary value, or when the estimated LET threshold is uncertain, the system enters the laser test path. When the estimated LET threshold is less than the preset LET threshold boundary value, the system enters the proton test path.

[0011] Step 3: Conducting the Laser Test Path

[0012] By using pulsed lasers to scan under typical operating conditions of AI chips, the type, location, and minimum trigger energy of single-particle events are determined and recorded, and the distribution of sensitive points is obtained.

[0013] Step 4: Processing laser test data

[0014] By mapping the minimum trigger energy to an equivalent linear energy transfer threshold, information on the security boundary and sensitive areas of the AI ​​chip can be obtained.

[0015] Step 5: Determining the Secondary Test Path

[0016] Determine whether the equivalent linear energy transfer threshold is greater than the preset LET threshold boundary. If yes, proceed to step nine; otherwise, proceed to step six.

[0017] Step Six: Implementation of the Proton Experiment Path

[0018] The AI ​​chip is irradiated with a proton beam, and the occurrence of single-event effects is statistically analyzed according to anomaly judgment rules to obtain the event occurrence pattern;

[0019] Step 7: Determination of Proton Test Results

[0020] If the anomaly decision rule determines that a single-event effect has occurred, proceed to step eight to collect the results; if the anomaly decision rule determines that no single-event effect has occurred, return to step three to conduct supplementary experiments.

[0021] Step 8: Processing Proton Test Data

[0022] Collect and record the phenotype, cross-section, and required fluence information of single-event effects in proton beam experiments;

[0023] Step 9: Select heavy ion test parameters

[0024] Based on the equivalent linear energy transfer threshold and flux information, select appropriate heavy ion species, incident energy, linear energy transfer value and flux step;

[0025] Step 10: Perform heavy ion experiments and confirm results

[0026] Conduct heavy ion experiments and compare the results with those from laser and proton experiments. When the equivalent threshold and phenotype are consistent within the preset tolerance, output the final evaluation conclusion.

[0027] A further improvement is that, in step one, the anomaly judgment rule is the criterion for determining whether an event is a valid single-event effect in a single-event test.

[0028] A further improvement is made in step two, where the preset LET threshold value is 10 MeV·cm²·mg. -1 .

[0029] Further improvements are made in the following way: In step three, the pulsed laser test is performed by adjusting the laser focus position to the sensitive area of ​​the AI ​​chip using a three-dimensional moving stage, adopting an energy step scanning method, and using wafer thinning, laser windowing, or two-photon absorption methods to ensure the effectiveness of laser incidence in cases of back-side incidence or metal obstruction.

[0030] The further improvement lies in the following: In step six, the specific method for performing the proton beam experiment is as follows: fix the AI ​​chip in the irradiation area, position the center of the proton beam at the geometric center of the chip, ensure uniform irradiation by adjusting the shielding, and select representative proton energies and fluxes for irradiation.

[0031] A further improvement is made in step nine, where, when selecting heavy ion test parameters, priority is given to heavy ion species whose range in silicon material can cover the physical thickness of the AI ​​chip.

[0032] A further improvement is made in step nine, where the range of the linear energy transfer value of the heavy ions is within the range of laser LET. eq,laser ±20% or 10 MeV·cm²·mg -1 Ion sources within [the specified range].

[0033] A further improvement is made in step ten: if the deviation between the heavy ion test result and the laser or proton test result exceeds the preset tolerance, and after excluding accidental factors, the heavy ion test result is used as the basis for the final evaluation conclusion.

[0034] The beneficial effects of this invention are as follows:

[0035] This invention uses a given linear energy transfer threshold as a boundary, employing pulsed lasers to rapidly assess high linear energy transfer conditions and locate sensitive paths, and proton beams to perform representative measurements of low linear energy transfer conditions. Equivalent calibration and minimal heavy-ion experimental point calibration verification are then performed. This approach reduces dependence on heavy-ion testing time, lowers experimental costs, simplifies organizational complexity, and shortens the evaluation cycle, while ensuring comparability and reproducibility of results. Simultaneously, it improves the spatial accuracy of fault location and outputs thresholds and occurrence patterns suitable for engineering decisions. This is because the advantages of lasers in forming high-density charge carriers within a small volume are utilized for high linear energy transfer scenarios, while the stable flux and close proximity to the shielded environment of proton beams are used for low linear energy transfer scenarios. Threshold demarcation and equivalent alignment reduce the incomparability caused by physical differences between the two methods. A complete experimental result closed loop is constructed using a small number of heavy-ion verifications to comprehensively evaluate the single-event effect of the device. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the steps of the present invention.

[0037] Figure 2 This is a flowchart illustrating the present invention.

[0038] Figure 3 This is a schematic diagram of the pulsed laser simulation single-event effect experimental device of the present invention.

[0039] Figure 4This is a schematic diagram of the proton and heavy ion irradiation test apparatus of the present invention.

[0040] The components include: 11. Single-particle testing system; 12. Pulsed laser; 13. Laser energy meter; 14. CCD imaging system; 15. Focusing objective lens; 16. Three-dimensional moving stage; 17. Sample to be tested; 18. Optical path system; 19. High-energy particle irradiation source; 20. Power supply module; 21. Main control system including computer; 22. Communication line; 23. Remote control computer; 24. Dedicated test board; 25. Irradiation area; 26. Irradiation chamber; 27. Measurement hall. Detailed Implementation

[0041] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0042] Example 1

[0043] according to Figure 1 As shown, this embodiment proposes a method for evaluating single-event effects in AI chips, including the following steps:

[0044] Step 1: Estimate the linear energy transfer threshold for single-event effects in AI chips.

[0045] Based on the layout and process data of the AI ​​chip, typical workload, existing similar sample data or related tests, and combined with the charge collection model, the linear energy transfer threshold required for the AI ​​chip under test to produce a single-event effect is determined, and an estimated LET threshold and anomaly judgment rule are formed. The anomaly judgment rule is the criterion for judging whether an event is a valid single-event effect in the single-event effect test.

[0046] Step 2: Preliminary test path selection

[0047] A preset LET threshold cutoff value is established, and then the estimated LET threshold is compared with the preset LET threshold cutoff value. If the estimated LET threshold is greater than or equal to the preset LET threshold cutoff value, or if the estimated LET threshold is uncertain, the laser test path is entered. If the estimated LET threshold is less than the preset LET threshold cutoff value, the proton test path is entered. The preset LET threshold cutoff value is 10 MeV·cm²·mg. -1 ;

[0048] Step 3: Conducting the Laser Test Path

[0049] Using pulsed lasers, scanning is performed on AI chips under typical operating conditions (depending on the specific device's datasheet) to determine and record the type, location, and minimum trigger energy of single-particle events, thereby obtaining the distribution of sensitive points. The pulsed laser experiment is conducted by adjusting the laser focus position to the sensitive area of ​​the AI ​​chip using a three-dimensional moving stage, employing an energy step scanning method, and using wafer thinning, laser windowing, or two-photon absorption methods to ensure the effectiveness of laser incidence in cases of back-side incidence or metal obstruction.

[0050] Step 4: Processing laser test data

[0051] By mapping the minimum trigger energy to an equivalent linear energy transfer threshold, information on the security boundary and sensitive areas of the AI ​​chip can be obtained.

[0052] Step 5: Determining the Secondary Test Path

[0053] Determine whether the equivalent linear energy transfer threshold is greater than the preset LET threshold boundary. If yes, proceed to step nine; otherwise, proceed to step six.

[0054] Step Six: Implementation of the Proton Experiment Path

[0055] The AI ​​chip is irradiated with a proton beam, and the occurrence of single-event effects is statistically analyzed according to the anomaly judgment rules to obtain the event occurrence pattern. The specific method of performing the proton beam experiment is as follows: the AI ​​chip is fixed in the irradiation area, the center of the proton beam is positioned at the geometric center of the chip, the shielding is adjusted to ensure uniform irradiation, and representative proton energies and fluxes are selected for irradiation.

[0056] Step 7: Determination of Proton Test Results

[0057] If the anomaly decision rule determines that a single-event effect has occurred, proceed to step eight to collect the results; if the anomaly decision rule determines that no single-event effect has occurred, return to step three to conduct supplementary experiments.

[0058] Step 8: Processing Proton Test Data

[0059] Collect and record the phenotype, cross-section, and required fluence information of single-event effects in proton beam experiments;

[0060] Step 9: Select heavy ion test parameters

[0061] Based on the equivalent linear energy transfer threshold and fluence information, appropriate heavy ion species, incident energy, linear energy transfer value, and fluence step are selected. The selection range for the linear energy transfer value of heavy ions is within the range of laser LET. eq,laser ±20% or 10 MeV·cm²·mg -1 For ion sources within the specified range, priority should be given to heavy ion species in silicon materials whose range can cover the physical thickness of the AI ​​chip.

[0062] Step 10: Perform heavy ion experiments and confirm results

[0063] Heavy ion experiments are conducted, and the results are compared and analyzed with those of laser and proton experiments. When the equivalent threshold and phenotype are consistent within the preset tolerance, the final evaluation conclusion is output. If the deviation between the heavy ion experiment results and the laser or proton experiment results exceeds the preset tolerance, and after excluding accidental factors, the heavy ion experiment results are used as the basis for the final evaluation conclusion.

[0064] Example 2

[0065] according to Figures 2-4 As shown, this embodiment proposes a method for evaluating single-event effects in AI chips, including the following steps:

[0066] Step 1: Estimate the linear energy transfer threshold for single-event effects in AI chips.

[0067] Based on the layout and process data of AI chips, typical workloads, existing similar sample data or related tests, the linear energy transfer threshold required for the device under test to generate a single-event response is estimated using a comprehensive charge collection model, forming an estimated threshold and anomaly judgment rules. These rules serve as the criteria for determining whether an event is a valid single-event effect in single-event effect testing.

[0068] Step 2: Preliminary test path selection

[0069] The estimated LET threshold will be compared with 10 MeV·cm²·mg -1 The comparison is performed; values ​​"≥10" or "threshold uncertain" enter the laser branch (step three), and values ​​"<10" enter the proton branch (step six). (10 MeV·cm²·mg) -1 As a boundary, since the evaluation of single-event effects of devices using protons should be based on the analysis of secondary particles generated by nuclear reactions as the dominant contribution, in engineering, the LET spectrum of this energy region in silicon is usually taken as 8 MeV·cm²·mg. -1 As an effective upper limit; while pulsed lasers can generate high-density charge carriers in a small volume, they are more suitable for high linear energy transmission situations above this limit; selecting the boundary threshold point can clearly define the division of labor while ensuring comparability, and reduce invalid machine time and repeated verification.

[0070] Step 3: Conducting the Laser Test Path

[0071] like Figure 3As shown, a laser experiment was conducted. Under representative working conditions of the sample 17, a pulsed laser experiment was carried out. By adjusting the three-dimensional moving stage 16 and observing the imaging of the CCD imaging system 14, it was determined that the laser focus fell on the sensitive area and formed a secondary spot. The start and end points of the scan were set, and an appropriate energy was selected for scanning. The event type, location, and minimum trigger energy were recorded to form the distribution of sensitive points and hot spots. For back-incidence or thick metal obstruction, thinning, windowing, or two-photon methods were used as needed to ensure the effectiveness of the incident light.

[0072] Step 4: Processing laser test data

[0073] Under uniform operating conditions, the minimum trigger energy is mapped to the equivalent linear energy transfer threshold (LET). eq,laser ±20% is used to obtain the sample safety boundary, and then current limiting, reset and other protective measures are specified, and information such as equivalent threshold, required injection volume and sensitive area points that are prone to triggering single-event effects is given;

[0074] Step 5: Determining the Secondary Test Path

[0075] Determine the single-event threshold (LET) obtained from laser experiments. eq,laser Is it ≥10 MeV·cm²·mg? -1 If the value is greater than the limit, proceed to step nine; if the value is lower than the limit, it indicates that the assumption that the LET threshold is in the high range is not valid, and proceed to the proton branch (step six) to determine the phenotype and cross-section of single-event effects occurring in close-to-real space.

[0076] Step Six: Implementation of the Proton Experiment Path

[0077] like Figure 4 As shown, the flux rate, energy, and other parameters of the proton accelerator were adjusted on-site. After fixing the sample to be tested in the irradiation area, the beam center was positioned at the geometric center of the chip under test, and the shielding was adjusted to ensure that the entire chip under test could be irradiated by protons. Representative proton energies and fluxes were selected for uniform irradiation. A main control system 21 containing a computer was used to statistically analyze the occurrence of single-event effects and the running time to obtain the event occurrence patterns.

[0078] Step 7: Determination of Proton Test Results

[0079] If a single-event effect occurs, the threshold for judgment is <10 MeV·cm²·mg. -1 If the device is sensitive, the low range can dominate the induction of single event (SEE), and proceed to step eight for statistical results; if no single event occurs, in order to cover the high LET irradiation situation, return to step three to supplement the laser test;

[0080] Step 8: Processing Proton Test Data

[0081] Under uniform conditions, the phenotype and calculation cross-section of single-event effects are recorded, and information such as the upper limit of the threshold, cross-section, phenotype and required injection volume for single-event effects when the actual space environment is closely approximated is given.

[0082] Step 9: Select heavy ion test parameters

[0083] Based on the equivalent linear energy transfer and flux obtained above, select suitable ion type, incident energy, linear energy transfer value, and flux step. The range for the heavy ion linear energy transfer value is within the range of laser LET. eq,laser ±20% or 10MeV·cm²·mg -1 For ion sources within the range, priority is given to ions in silicon whose range covers the thickness of the chip (such as Kr and Ar ions). The principle of "minimizing point calibration" is followed to determine a small number of heavy ion test verification points to save valuable machine time. At the same time, power supply current limit and monitoring threshold are set to ensure sample safety.

[0084] Step 10: Perform heavy ion experiments and confirm results

[0085] Compare and analyze the results with those of laser and proton experiments; when the equivalent threshold and phenotype are consistent within the preset tolerance, output the final conclusion and evidence chain; if the deviation exceeds the limit, the heavy ion experiment results shall prevail under the condition that the non-accidental factors are ensured.

[0086] Example 3

[0087] like Figure 3 As shown, this embodiment provides a pulsed laser simulation device for single-event effects, which includes a single-event testing system 11, a pulsed laser 12, a laser energy meter 13, a CCD imaging system 14, a focusing objective lens 15, a three-dimensional moving stage 16, a sample to be tested 17, and an optical path system 18. Its core principle is to utilize the laser pulse generated by the pulsed laser 12, which is guided and adjusted by the optical path system 18, and finally focused by the focusing objective lens 15 onto the sample to be tested 17 placed on the three-dimensional moving stage 16, to simulate the single-event effect generated by the incident high-energy particles.

[0088] Correspondingly, the single-event testing system 11 is the brain of the device, consisting of a computer, test board and dedicated software. It controls the output of the pulsed laser 12 and the movement of the three-dimensional moving stage 16 for scanning. At the same time, it receives electrical signals from the sample 17 under test, monitors whether a single-event effect occurs under laser pulse bombardment, and records the occurrence of the event.

[0089] The pulsed laser 12 is the excitation source of the device. The laser it generates first enters the optical path system 18. The optical path system 18 divides the laser beam into two paths and guides them. The first path is the main optical path, in which the laser pulse continues to be transmitted to the focusing objective lens 15. The second path is the monitoring optical path, in which a portion of the laser energy is guided to the laser energy meter 13 for real-time monitoring and calibration of the laser pulse energy, ensuring the accuracy and repeatability of the experimental conditions.

[0090] Furthermore, the sample 17 to be tested is fixed on a three-dimensional moving stage 16, which can perform precise three-dimensional movement to position the laser focus on any sensitive area of ​​the chip. The CCD imaging system 14 and the optical path system 18 are integrated through a common optical path design to observe the surface of the sample to be tested in real time before and during the test, assisting in precise positioning.

[0091] Example 4

[0092] like Figure 4 As shown, this embodiment provides a pulsed laser simulation device for single-event effects, which includes a high-energy particle irradiation source 19, a power supply module 20, a main control system 21 containing a computer, a communication line 22, a remote control computer 23, a dedicated test board 24, an irradiation area 25, an irradiation chamber 26, and a measurement hall 27. The core function of the device is to use the proton beam or heavy ion beam generated by the high-energy particle irradiation source 19 to irradiate the sample 17 (AI chip) mounted on the dedicated test board 24 in the irradiation area 25 to simulate the space radiation environment and remotely monitor its single-event effects.

[0093] Specifically, Measurement Hall 27 is a safe area for personnel to conduct remote operation and monitoring. Irradiation Chamber 26 contains Irradiation Zone 25, which is the core and radioactive hazardous area for high-energy particle irradiation.

[0094] The high-energy particle irradiation source 19 is powered by the power supply module 20, which generates a collimated proton or heavy ion beam that directly irradiates the dedicated test plate 24 located within the irradiation zone 25. Inside the irradiation chamber 26, the dedicated test plate 24 is connected via cables to a main control system 21 containing a computer. This main control system 21 is responsible for powering the sample under test, loading the workload, and monitoring and recording the single-event effects occurring on the sample during irradiation in real time.

[0095] To isolate personnel from the radiation environment, the main control system 21 is connected to a remote control computer 23 installed in the measurement hall 27 via a communication line 22 (which must meet the safe distance requirement between the irradiation room 26 and the measurement hall 27). Within this safe area, experimental personnel send commands (such as start, stop, and parameter adjustment) to the main control system 21 via the remote computer 23 and receive real-time test data transmitted back by the main control system 21.

[0096] 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 present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating single-event effects in an AI chip, characterized in that: Includes the following steps: Step 1: Estimate the linear energy transfer threshold for single-event effects in AI chips. Based on the layout and process data of the AI ​​chip, typical workload, existing similar sample data or related tests, and combined with the charge collection model, the linear energy transfer threshold required for the AI ​​chip under test to produce a single-event effect is determined, and an estimated LET threshold and anomaly judgment rules are formed. Step 2: Preliminary test path selection The system sets a preset LET threshold boundary value, and then compares the estimated LET threshold with the preset LET threshold boundary value. When the estimated LET threshold is greater than or equal to the preset LET threshold boundary value, or when the estimated LET threshold is uncertain, the system enters the laser test path. When the estimated LET threshold is less than the preset LET threshold boundary value, the system enters the proton test path. Step 3: Conducting the Laser Test Path By using pulsed lasers to scan under typical operating conditions of AI chips, the type, location, and minimum trigger energy of single-particle events are determined and recorded, and the distribution of sensitive points is obtained. Step 4: Processing laser test data By mapping the minimum trigger energy to an equivalent linear energy transfer threshold, information on the security boundary and sensitive areas of the AI ​​chip can be obtained. Step 5: Determining the Secondary Test Path Determine whether the equivalent linear energy transfer threshold is greater than the preset LET threshold boundary. If yes, proceed to step nine; otherwise, proceed to step six. Step Six: Implementation of the Proton Experiment Path The AI ​​chip is irradiated with a proton beam, and the occurrence of single-event effects is statistically analyzed according to anomaly judgment rules to obtain the event occurrence pattern; Step 7: Determination of Proton Test Results If the anomaly decision rule determines that a single-event effect has occurred, proceed to step eight to collect the results; if the anomaly decision rule determines that no single-event effect has occurred, return to step three to conduct supplementary experiments. Step 8: Processing Proton Test Data Collect and record the phenotype, cross-section, and required fluence information of single-event effects in proton beam experiments; Step 9: Select heavy ion test parameters Based on the equivalent linear energy transfer threshold and flux information, select appropriate heavy ion species, incident energy, linear energy transfer value and flux step; Step 10: Perform heavy ion experiments and confirm results Conduct heavy ion experiments and compare the results with those from laser and proton experiments. When the equivalent linear energy transfer threshold and phenotype are consistent within the preset tolerance, output the final evaluation conclusion.

2. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step one, the anomaly judgment rule is the criterion for determining whether an event is a valid single-event effect in a single-event test.

3. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step two, the preset LET threshold threshold value is 10 MeV·cm²·mg. -1 .

4. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step three, the pulsed laser test is performed by adjusting the laser focus position to the sensitive area of ​​the AI ​​chip using a three-dimensional moving stage, employing an energy step scanning method, and using wafer thinning, laser windowing, or two-photon absorption methods to ensure the effectiveness of laser incidence in cases of back-side incidence or metal obstruction.

5. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step six, the specific method for performing the proton beam experiment is as follows: fix the AI ​​chip in the irradiation area, position the center of the proton beam at the geometric center of the AI ​​chip, ensure uniform irradiation by adjusting the shielding, and select representative proton energies and fluxes for irradiation.

6. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step nine, when selecting heavy ion test parameters, priority should be given to heavy ion species whose range in silicon material can cover the physical thickness of the AI ​​chip.

7. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step nine, the range of the linear energy transfer value of heavy ions is determined by the laser LET. eq,laser ±20% or 10 MeV·cm²·mg -1 Ion sources within [the specified range].

8. The single-event effect evaluation method for an AI chip according to claim 1, characterized in that: In step ten, if the deviation between the heavy ion test result and the laser or proton test result exceeds the preset tolerance, and after excluding accidental factors, the heavy ion test result shall be used as the basis for the final evaluation conclusion.

Images