Simulation integrated circuit test method based on edge calculation

Abstract

The invention discloses an analog integrated circuit test method based on edge calculation, the method is executed by an edge node, and the method comprises the following steps: obtaining a step response waveform of an analog integrated circuit; extracting screening features from the step response waveform; determining an abnormal level and a test type of the screening feature; determining the demand level of the server for the original waveform data based on the abnormal level of the screening feature and the test type; and determining a sending strategy for sending the screening feature and the original waveform data to the server based on the demand level. According to the invention, waiting and repeated interaction of the server can be reduced, and test judgment efficiency and test throughput are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for testing analog integrated circuits based on edge computing. Background Technology

[0002] Analog integrated circuit testing is a crucial component of integrated circuit testing, widely used for performance testing and quality screening of operational amplifiers, low-dropout linear regulators, analog-to-digital converters, digital-to-analog converters, power management chips, and various analog front-end circuits. As the complexity of analog integrated circuits increases and testing throughput requirements rise, the traditional method of centralized test data analysis and result determination by a central server is gradually revealing problems such as high latency, large bandwidth consumption, and insufficient scalability. Therefore, edge computing is increasingly being introduced into analog integrated circuit testing scenarios. In an edge computing-based testing architecture, edge nodes located close to the test equipment can perform preprocessing, feature extraction, compression and encapsulation, and preliminary analysis of the raw waveforms, while the server handles more complex model inference, comprehensive judgment, and strategy management, thereby reducing the amount of raw data transmitted back and improving test real-time performance.

[0003] In existing technologies, edge nodes typically extract feature data for rapid screening from raw test data and send it to the server first; while the large amount of raw key waveform data is often cached, queued for compression, or sent later. However, in some scenarios, the server cannot accurately determine the characteristics of the target circuit based solely on screening features and still needs to perform further analysis in conjunction with the corresponding raw key waveform. If the raw key waveform has not yet been compressed or is still in the edge node's queue, the server needs to wait for subsequent uploads or initiate a supplementary request again. Therefore, in existing technologies, screening feature data and raw key waveform data are usually processed separately, and the upload sequence is mainly determined based on data volume or default rules, failing to effectively coordinate with the actual judgment requirements on the server side. This easily leads to secondary waiting by the server, increased cloud-edge interaction rounds, difficulty in completing the analysis process in one closed loop, and lengthens the overall test cycle, affecting the testing efficiency of analog integrated circuits. Summary of the Invention

[0004] This invention acquires the step response waveform of an analog integrated circuit through edge nodes, extracts screening features, and determines the server's demand level for the raw waveform data based on the anomaly level and test type, thereby selecting an appropriate transmission strategy. For high-demand scenarios, the edge node sends screening features, associated descriptive information, and a key summary of the raw waveform data; for low-demand scenarios, only screening features and associated descriptive information are sent, while the raw waveform data is stored locally. This reduces server waiting and repetitive interactions, improving test judgment efficiency and test throughput.

[0005] This invention provides a method for testing analog integrated circuits based on edge computing. The method is executed by edge nodes and includes: Obtain the step response waveform of an analog integrated circuit; Extract screening features from the step response waveform; Determine the abnormality level of the screening features and the type of test; The server's demand level for raw waveform data is determined based on the anomaly level of the screening features and the test type; and The strategy for sending screening features and raw waveform data to the server is determined based on the demand level.

[0006] In a preferred embodiment, the screening features include peak value, establishment time, and anomaly score; The methods also include: Generate associated descriptive information related to the screening features. The associated descriptive information includes the test object identifier, timestamp, and priority identifier.

[0007] In a preferred embodiment, the demand levels include a first demand level and a second demand level; Among them, the strategy for determining the transmission of screening features and raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, associated description information, and key summaries of the raw waveform data to the server. If the demand level is the second demand level, the screening features and associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

[0008] In a preferred embodiment, the server is configured to perform the following operations: Determine whether the identification of analog integrated circuits can be completed based on screening features; If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then continue to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

[0009] In a preferred embodiment, the server is also configured to perform the following operations: If a key summary of the received raw waveform data is determined, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the raw waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the raw waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the identification of analog integrated circuits cannot be completed based on screening features and key summaries of raw waveform data, the remaining data of the raw waveform data is requested from the edge node based on the associated description information.

[0010] This invention provides an edge computing-based analog integrated circuit testing system, comprising edge nodes and a server, wherein the edge nodes are configured to perform the following operations: Obtain the step response waveform of an analog integrated circuit; Extract screening features from the step response waveform; Determine the abnormality level of the screening features and the type of test; The server's demand level for raw waveform data is determined based on the anomaly level of the screening features and the test type; and The strategy for sending screening features and raw waveform data to the server is determined based on the demand level.

[0011] In a preferred embodiment, the screening features include peak value, establishment time, and anomaly score; The edge nodes are also configured to perform the following operations: Generate associated descriptive information related to the screening features. The associated descriptive information includes the test object identifier, timestamp, and priority identifier.

[0012] In a preferred embodiment, the demand levels include a first demand level and a second demand level; Among them, the strategy for determining the transmission of screening features and raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, associated description information, and key summaries of the raw waveform data to the server. If the demand level is the second demand level, the screening features and associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

[0013] In a preferred embodiment, the server is configured to perform the following operations: Determine whether the identification of analog integrated circuits can be completed based on screening features; If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then continue to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

[0014] In a preferred embodiment, the server is also configured to perform the following operations: If a key summary of the received raw waveform data is determined, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the raw waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the raw waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the identification of analog integrated circuits cannot be completed based on screening features and key summaries of raw waveform data, the remaining data of the raw waveform data is requested from the edge node based on the associated description information.

[0015] Compared with existing technologies, this invention has the following advantages: In high-demand scenarios, edge nodes send screening features, associated description information, and key summaries of the original waveform data; in low-demand scenarios, only screening features and associated description information are sent, and the original waveform data is saved locally. This reduces server waiting and repetitive interactions, improving test judgment efficiency and test throughput. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system architecture of one embodiment of the present invention.

[0017] Figure 2 This is a logic block diagram of one embodiment of the present invention.

[0018] Figure 3 This is a flowchart of a method according to an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the load step response waveform according to an embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0021] like Figure 1As shown, the system of the present invention may include an edge node 1100, a server 1200, and a simulated integrated circuit under test 1300. The edge node 1100 is communicatively connected to the simulated integrated circuit under test 1300 and is used to acquire the step response waveform of the simulated integrated circuit under test 1300 and extract screening features from the step response waveform. Screening features may include peak value, setup time, and anomaly score. The edge node 1100 may also determine the server 1200's demand level for raw waveform data based on the anomaly level of the screening features and the test type, and further determine a corresponding transmission strategy based on the demand level. The server 1200 is communicatively connected to the edge node 1100 and is used to receive at least one of the screening features, associated description information, and key summaries of the raw waveform data sent by the edge node 1100, and perform judgment processing on the simulated integrated circuit under test 1300 based on the received data. In some embodiments, when the server 1200 determines that the judgment of the simulated integrated circuit under test 1300 cannot be completed based solely on the screening features, it may further request the raw waveform data or the remaining data of the raw waveform data from the edge node 1100 based on the associated description information. Therefore, through the collaborative processing between edge node 1100 and server 1200, the waiting time for raw waveform data during the judgment process of server 1200 can be reduced, thereby improving the testing efficiency of the analog integrated circuit 1300 under test. like Figure 2 As shown, the system of the present invention may include an edge node 2100, a server 2200, and a simulated integrated circuit under test 2300. The edge node 2100 may include a waveform acquisition unit 2110, a feature extraction unit 2120, a level determination unit 2130, an association information generation unit 2140, a transmission strategy determination unit 2150, a first transmission unit 2160, a second transmission unit 2170, and a local storage unit 2180. The waveform acquisition unit 2110 is used to acquire the step response waveform of the analog integrated circuit 2300 under test; the feature extraction unit 2120 is used to extract screening features from the step response waveform, which may include peak value, settling time, and anomaly score; the level determination unit 2130 is used to determine the anomaly level and test type of the screening features, and determine the server 2200's demand level for the original waveform data based on the anomaly level and test type; the association information generation unit 2140 is used to generate association description information related to the screening features, which may include test object identifier, timestamp, and priority identifier; and the sending strategy determination unit 2150 is used to determine the sending strategy for sending the screening features and original waveform data to the server 2200 based on the demand level.

[0022] In some implementations, when the level determination unit 2130 determines the demand level to be the first demand level, the sending strategy determination unit 2150 controls the first sending unit 2160 to send the screening features, associated description information, and key summaries of the original waveform data to the server 2200; when the level determination unit 2130 determines the demand level to be the second demand level, the sending strategy determination unit 2150 controls the second sending unit 2170 to send the screening features and associated description information to the server 2200, and saves the original waveform data in the local storage unit 2180.

[0023] Server 2200 may include an initial determination unit 2210, a summary determination unit 2220, and a data request unit 2230. The initial determination unit 2210 determines whether the analog integrated circuit 2300 can be determined based on screening features; if the determination can be completed, a determination result is output; if the determination cannot be completed, it further determines whether a key summary of the original waveform data has been received. If no key summary is received, the data request unit 2230 may request the original waveform data from the edge node 2100 based on the associated description information. If a key summary has been received, the summary determination unit 2220 determines whether the analog integrated circuit 2300 can be determined based on the screening features and the key summary; if the determination can be completed, a determination result is output; if the determination still cannot be completed, the data request unit 2230 further requests the remaining data of the original waveform data from the edge node 2100 based on the associated description information.

[0024] like Figure 3 As shown, the method of the present invention includes the following steps: Step 1: Obtain the step response waveform of the analog integrated circuit; In a specific example, the step response waveform can be a load step response waveform; an example of a load step response waveform can be found here. Figure 4 The load step response waveform is generally used to evaluate the dynamic regulation capability of circuits such as LDOs, regulators, and reference sources. When the load current changes abruptly, the output voltage will drop or overshoot, and then gradually recover to a steady state. This waveform is mainly used to analyze transient response capability, recovery speed, and loop stability.

[0025] Step 2: Extract screening features from the step response waveform; In specific examples, with Figure 4Taking the load step response waveform as an example, the peak value can be understood as the extreme point after the step event. Since the load step response waveform first exhibits an undershoot after the load change and then an overshoot during the recovery process, two relevant extreme values ​​can be extracted from the waveform. Specifically, the undershoot peak value, i.e., the lowest point of the output voltage, approximately 0.99V, can be extracted; the overshoot peak value, i.e., the highest point of the output voltage, approximately 1.28V, can also be extracted. In some implementations, the peak value is preferably defined as the abnormal main peak value after the step. In this example, the abnormal main peak value can usually be taken as the lowest point of 0.99V, because this lowest point can more directly reflect the output voltage drop caused by the load change.

[0026] Settlement time can be defined as the time elapsed from the moment a load step occurs until the output voltage re-enters the target steady-state range and remains within that range. According to... Figure 4 The load step occurs approximately 20µs. If a steady-state target value of 1.2V is used as the reference, and a ±2% steady-state error band is adopted, the corresponding allowable range is 1.176V to 1.224V, with an upper boundary of 1.224V and a lower boundary of 1.176V. Based on... Figure 4 The output voltage enters the error band and remains stable at approximately 37.5µs. Therefore, the settling time corresponding to this graph can be approximated as 37.5µs minus 20µs, which is approximately 17.5µs. Furthermore, if a more stringent ±1% error band is used, the settling time will be longer, approximately 25.1µs.

[0027] In a specific example, multiple waveform anomaly indicators can be normalized before being weighted and summed to obtain the anomaly score. For Figure 4 The load step response waveform shown can be used to construct anomaly scores using at least the following three quantities. First, the voltage drop amplitude, which reflects the maximum deviation of the output voltage from its steady-state value after a sudden load change. Figure 4 For example, when the steady-state reference value is 1.2V and the lowest point is 0.99V, the voltage drop can be taken as 0.21V. Second, settling time. A longer settling time generally indicates a slower output recovery, and correspondingly, the anomaly may be more pronounced. Based on the aforementioned waveform example, the settling time can be taken as approximately 17.5µs. Third, overshoot amplitude, used to characterize the maximum deviation exceeding the steady-state value during the recovery process. Figure 4 For example, when the highest point is 1.28V and the steady-state reference value is 1.2V, the overshoot amplitude can be taken as 0.08V.

[0028] Furthermore, reference thresholds can be set for the three quantities mentioned above. For example, the reference threshold for voltage drop amplitude can be set to 0.20V, the reference threshold for settling time to 20µs, and the reference threshold for overshoot amplitude to 0.10V. For any waveform to be analyzed, the actual measured voltage drop amplitude, settling time, and overshoot amplitude can be compared with their corresponding reference thresholds to obtain their respective normalized results. To avoid a single indicator being too large and having an excessive impact on the overall result, the normalized results can be limited to between 0 and 1. That is, when an actual value is greater than its corresponding reference threshold, the normalized result can be recorded as 1; when an actual value is less than its corresponding reference threshold, the normalized result can be determined according to its ratio to the reference threshold. After completing the normalization process, the normalized results can be weighted and summed to obtain an anomaly score. In some implementations, the weight corresponding to voltage drop amplitude can be set to 0.4, the weight corresponding to settling time can be set to 0.4, and the weight corresponding to overshoot amplitude can be set to 0.2. According to this setting, voltage drop amplitude and settling time have a relatively greater impact on the anomaly score, while overshoot amplitude has a relatively smaller impact. Furthermore, after obtaining the anomaly score, it can be classified into different levels based on preset thresholds. For example, the first threshold T1 can be set to 0.40, and the second threshold T2 can be set to 0.75. When the anomaly score is less than 0.40, it can be classified as a low anomaly; when the anomaly score is greater than or equal to 0.40 and less than 0.75, it can be classified as a medium anomaly; and when the anomaly score is greater than or equal to 0.75, it can be classified as a high anomaly.

[0029] The following explanation is based on specific numerical values.

[0030] In the first numerical example, assume the steady-state reference value of a load step response waveform is 1.2V, the lowest point is 0.95V, the highest point is 1.31V, and the settling time is 24µs. Therefore, the voltage drop amplitude of this waveform is 0.25V, the overshoot amplitude is 0.11V, and the settling time is 24µs. Since 0.25V is greater than the voltage drop amplitude reference threshold of 0.20V, the normalization result for this term can be taken as 1; since 24µs is greater than the settling time reference threshold of 20µs, the normalization result for this term can also be taken as 1; since 0.11V is greater than the overshoot amplitude reference threshold of 0.10V, the normalization result for this term can also be taken as 1. After weighting according to the aforementioned weights, the anomaly score is 1.00. Since 1.00 is greater than the second threshold of 0.75, this example can be judged as a high anomaly.

[0031] In the second numerical example, assume the steady-state reference value of a load step response waveform is still 1.2V, the lowest point is 1.03V, the highest point is 1.25V, and the settling time is 18µs. Therefore, the voltage drop amplitude of this waveform is 0.17V, the overshoot amplitude is 0.05V, and the settling time is 18µs. Comparing each actual value with the corresponding reference threshold yields the following normalization results: the normalization result for the voltage drop amplitude is approximately 0.85, because the ratio of 0.17V to 0.20V is approximately 0.85; the normalization result for the settling time is approximately 0.90, because the ratio of 18µs to 20µs is approximately 0.90; and the normalization result for the overshoot amplitude is approximately 0.50, because the ratio of 0.05V to 0.10V is approximately 0.50. After weighting according to the aforementioned weights, the anomaly score is approximately 0.80. Since 0.80 is greater than the second threshold of 0.75, this example can also be classified as a high anomaly.

[0032] In the third example, assume the steady-state reference value of a load step response waveform is still 1.2V, the lowest point is 1.08V, the highest point is 1.24V, and the settling time is 13µs. Therefore, the voltage drop amplitude of this waveform is 0.12V, the overshoot amplitude is 0.04V, and the settling time is 13µs. Further, the normalized result corresponding to the voltage drop amplitude can be approximately 0.60, the normalized result corresponding to the settling time can be approximately 0.65, and the normalized result corresponding to the overshoot amplitude can be approximately 0.40. After weighting according to the aforementioned weights, the anomaly score can be approximately 0.58. Since 0.58 is greater than or equal to the first threshold of 0.40 and less than the second threshold of 0.75, this example can be judged as a medium anomaly.

[0033] In the fourth example, assume the steady-state reference value of a load step response waveform is 1.2V, the lowest point is 1.14V, the highest point is 1.22V, and the settling time is 8µs. Therefore, the voltage drop amplitude of this waveform is 0.06V, the overshoot amplitude is 0.02V, and the settling time is 8µs. Further, the normalized result corresponding to the voltage drop amplitude can be approximately 0.30, the normalized result corresponding to the settling time can be approximately 0.40, and the normalized result corresponding to the overshoot amplitude can be approximately 0.20. After weighting according to the aforementioned weights, the anomaly score can be approximately 0.32. Since 0.32 is less than the first threshold of 0.40, this example can be judged as a low anomaly.

[0034] Step 3: Determine the anomaly level of the screening features and the test type; In a specific example, the anomaly level can be directly determined based on the aforementioned anomaly score. For instance, the aforementioned threshold setting method can be continued, setting the first threshold T1 to 0.40 and the second threshold T2 to 0.75. When the anomaly score is less than 0.40, it can be determined as a low anomaly; when the anomaly score is greater than or equal to 0.40 and less than 0.75, it can be determined as a medium anomaly; and when the anomaly score is greater than or equal to 0.75, it can be determined as a high anomaly.

[0035] by Figure 4 Taking the load step response waveform shown as an example, the following screening characteristics have been obtained in the preceding steps: steady-state reference value is 1.2V, undershoot peak value is approximately 0.99V, overshoot peak value is approximately 1.28V, and settling time is approximately 17.5µs. In this example, the voltage drop amplitude can be taken as 0.21V, the overshoot amplitude can be taken as 0.08V, and the settling time can be taken as 17.5µs. If the aforementioned reference thresholds are used respectively, namely a voltage drop reference threshold of 0.20V, a setup time reference threshold of 20µs, and an overshoot reference threshold of 0.10V, the following normalization results can be obtained: Since 0.21V is greater than 0.20V, the normalization result corresponding to the voltage drop can be recorded as 1; since 17.5µs is less than 20µs, the normalization result corresponding to the setup time can be determined as approximately 0.875 based on the ratio of 17.5µs to 20µs; since 0.08V is less than 0.10V, the normalization result corresponding to the overshoot can be determined as approximately 0.80 based on the ratio of 0.08V to 0.10V. Continuing to use the aforementioned weighting method, i.e., a voltage drop weight of 0.4, a setup time weight of 0.4, and an overshoot weight of 0.2, the anomaly score of this waveform is approximately 0.91. Since 0.91 is greater than the second threshold of 0.75, the edge nodes can... Figure 4 The anomaly level of the corresponding test results is determined to be high anomaly.

[0036] In specific examples, the test type can be determined based on the test context information during waveform acquisition. If the waveform is acquired when the load current undergoes a sudden change, and the output voltage exhibits an initial drop or overshoot followed by a gradual recovery to a steady state, then the test type can be determined as a load step response test.

[0037] Continue with Figure 4 For example, if the testing device records that the load current switches from a first load value to a second load value, such as from 10mA to 100mA, when acquiring the waveform, and the output voltage shows a significant undershoot after the switch and gradually recovers, then the edge node can use this test context information to... Figure 4 The corresponding test type is determined to be a load step response test. In other words, for... Figure 4The edge node can not only be determined to be of high anomaly level, but also to be of load step response test type.

[0038] Step 4: Determine the server's requirement level for raw waveform data based on the anomaly level of the screening features and the test type; Step 5: Determine the sending strategy for sending screening features and raw waveform data to the server based on the demand level.

[0039] In a specific example, the first demand level can indicate that the server has a high demand for the original waveform data, and the edge node should send the screening features, associated description information, and key summaries of the original waveform data to the server; the second demand level can indicate that the server has a lower immediate demand for the original waveform data, and the edge node only needs to send the screening features and associated description information to the server and save the original waveform data locally.

[0040] In specific examples, the mapping relationship between exception levels and test types to requirement levels can be pre-defined. For example, the following mapping rule can be used: When the anomaly level is high, regardless of the step response test type, the server's demand level for the original waveform data can be determined as the first demand level. When the anomaly level is medium anomaly and the test type is load step response test, the demand level can also be determined as the first demand level. This is because load step response test usually focuses more on output voltage drop, recovery speed and loop stability. The server often needs to further examine the original waveform data or its key summaries to make a more accurate judgment. When the anomaly level is medium anomaly and the test type is other than load step response test, the requirement level can be determined as the second requirement level. When the anomaly level is low, the demand level can be set to the second demand level.

[0041] Continue with Figure 4 Taking the load step response waveform shown as an example, as determined in the preceding steps, the anomaly level of this waveform is high anomaly, and the test type is load step response test. In this case, the edge node can determine the server's demand level for the original waveform data corresponding to this waveform as the first demand level according to the preset mapping rules. In other words, for... Figure 4 The edge nodes not only send screening features such as peak values, setup time, and anomaly scores, as well as associated descriptive information, to the server, but can also send key summaries of the original waveform data to the server so that the server can complete the judgment of the analog integrated circuit more quickly.

[0042] In a preferred embodiment, the screening features include peak value, establishment time, and anomaly score; The methods also include: Generate associated descriptive information related to the screening features. The associated descriptive information includes the test object identifier, timestamp, and priority identifier.

[0043] In specific examples, the test object identifier can be used to characterize the test object to which the current screening feature belongs. In some implementations, the test object identifier can be a unique number of the simulated integrated circuit under test in the test system, such as a chip serial number, a combination of the test station number and the chip number, the test position number on the probe station, the device number in the sorting equipment, or other unique identification codes assigned by the test system. Through the test object identifier, the server can distinguish the screening features and raw waveform data corresponding to different simulated integrated circuits under test, avoiding data confusion between different devices in parallel or continuous testing scenarios.

[0044] Timestamps can be used to characterize the acquisition or generation time of the data corresponding to the current screening feature. In some implementations, a timestamp can be absolute time information recorded by an edge node when acquiring a step response waveform, or it can be relative time information relative to a certain test start time. For example, a timestamp can use an absolute time format such as "2026-03-28 10:15:23.125" or a relative time format such as "3rd round of testing, 20us after the start of the 2nd sampling window". Through timestamps, the server can further distinguish multiple sets of screening features and raw waveform data formed by the same test object in different test rounds or different sampling stages.

[0045] Priority identifiers can be used to characterize the importance or urgency of the raw waveform data corresponding to the current screening feature in subsequent server processing. In some implementations, priority identifiers can be represented in hierarchical form, such as high priority, medium priority, and low priority; or they can be represented in numerical form, such as priority 1, priority 2, and priority 3, where smaller values ​​represent higher priority, or larger values ​​represent higher priority. Preferably, priority identifiers can be further generated by edge nodes based on anomaly level, test type, or demand level. For example, when the anomaly level corresponding to the screening feature is high anomaly, or the server's demand level for the raw waveform data is first demand level, the corresponding priority identifier can be set to high priority; when the anomaly level is low, or the demand level is second demand level, the corresponding priority identifier can be set to low priority.

[0046] In a specific implementation, after receiving the screening features, associated description information, and key summaries or raw waveform data sent by the edge node, the server can establish an association between the screening features and the raw waveform data based on the test object identifier, timestamp, and priority identifier. Specifically, the server can first perform object-level matching on the received data based on the test object identifier to determine whether the screening features and raw waveform data belong to the same analog integrated circuit under test. After determining that they belong to the same test object, the server can further perform time matching on the data generated by the same test object in different test rounds or at different sampling times based on the timestamp to determine which time the raw waveform data corresponding to the screening feature was collected. After completing object matching and time matching, the server can also determine the processing order of the current screening feature and its corresponding raw waveform data based on the priority identifier, thereby prioritizing the processing of data with higher importance.

[0047] For example, in a multi-station parallel testing scenario, an edge node can generate a set of screening features for the first analog integrated circuit under test, attaching a test object identifier "DUT-01", a timestamp "10:15:23.125", and a priority identifier "high". Simultaneously, the edge node can generate another set of screening features for the second analog integrated circuit under test, attaching a test object identifier "DUT-02", a timestamp "10:15:23.240", and a priority identifier "medium". Upon receiving this data, the server first distinguishes "DUT-01" and "DUT-02" as different test objects based on the test object identifier. Then, based on the timestamp, it locates the corresponding original waveform data from their respective cache queue, key summary queue, or original waveform data queue. Finally, it prioritizes processing the data corresponding to "DUT-01" based on the priority identifier. Thus, the server can accurately associate the screening features with the original waveform data, improving the efficiency of subsequent judgment and processing.

[0048] For example, in scenarios where the same test object undergoes multiple rounds of testing, even if multiple screening features correspond to the same test object identifier, the server can still use timestamps to further distinguish the original waveform data corresponding to different rounds of testing. Suppose that the timestamp corresponding to the screening feature generated in the first round of testing for the same analog integrated circuit identified as "DUT-03" is "10:20:15.010", and the timestamp corresponding to the screening feature generated in the second round of testing is "10:20:16.580". Then, the server can associate the first round screening feature with the first round of original waveform data, and the second round screening feature with the second round of original waveform data, respectively, based on the timestamps, thereby avoiding incorrect matching of data generated by the same test object at different times.

[0049] In a preferred embodiment, the demand levels include a first demand level and a second demand level; Among them, the strategy for determining the transmission of screening features and raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, associated description information, and key summaries of the raw waveform data to the server. In a specific example, the key summary may include at least one of the following: a local sampling sequence of the downswing interval, a local sampling sequence of the recovery interval, a sampling sequence of the ringing interval before steady-state recovery, a set of waveform inflection point positions, segmented slope information, or a set of representative sampling points within a preset time window.

[0050] In a specific example, the edge node can first segment the original waveform data. For instance, the load step response waveform can be divided into an undershoot segment, a recovery segment, and a steady-state segment. The undershoot segment corresponds to the interval where the output voltage drops rapidly after a load change; the recovery segment corresponds to the interval where the output voltage gradually recovers from its lowest point, accompanied by possible overshoot or ringing; and the steady-state segment corresponds to the interval where the output voltage enters the vicinity of the steady-state error band. The edge node can select representative local waveform segments from the undershoot and recovery segments as key summaries, without having to send the entire original waveform data.

[0051] by Figure 4 Taking the load step response waveform shown as an example, the screening features extracted and sent to the server in the aforementioned steps may include: an abnormal main peak value of 0.99V, a setup time of approximately 17.5µs, and an anomaly score of approximately 0.91. Therefore, when constructing the key summary, the values ​​of 0.99V, 17.5µs, and 0.91 are no longer repeatedly sent; instead, values ​​that characterize... Figure 4 Local waveform information of waveform details.

[0052] In a specific example, it can be Figure 4The key summary is set into three parts. The first part is a representative set of sampling points in the undershoot interval. For example, several sampling points can be selected in the undershoot interval after a load step change. The time-voltage information corresponding to these sampling points may include: 20µs corresponds to 1.20V, 22µs corresponds to 1.08V, 24µs corresponds to 0.99V, and 26µs corresponds to 1.05V. This key summary can be used to characterize the local morphology of the output voltage rapidly dropping and reaching its lowest point after a load change. The second part is a representative set of sampling points in the recovery interval. For example, several sampling points can be further selected within the recovery interval. The time-voltage information corresponding to these sampling points may include: 28µs corresponds to 1.14V, 30µs corresponds to 1.23V, 32µs corresponds to 1.28V, 34µs corresponds to 1.24V, and 36µs corresponds to 1.20V. This key summary can be used to characterize the process of the output voltage recovering from its lowest point, experiencing overshoot, and gradually decaying. The third part is local morphology identification information. For example, a morphological marker can be added indicating "downshoot followed by recovery with one significant overshoot," or a structural marker can be added indicating "a single significant overshoot exists, but no sustained oscillation occurs." Through these three parts, the server can gain a relatively clear understanding of the waveform even without receiving the complete original waveform. Figure 4 The corresponding local change process of the waveform.

[0053] In other embodiments, the key summary may also take a more compact data form. For example, only a few key inflection points and their corresponding positions may be sent, instead of the complete local sampling sequence. (Continuing with...) Figure 4 For example, the key summary can be set as follows: the load mutation start point is approximately 20µs, the lowest point is approximately 24µs, the recovery overshoot peak is approximately 32µs, and the point where it reappears near the steady-state range is approximately 36µs. This can also be combined with local waveform trend information near each inflection point, such as "rapid decline in the 20µs to 24µs range," "resumption of rise in the 24µs to 32µs range," and "falling back to near the steady state range in the 32µs to 36µs range." Although this method does not directly send the complete waveform, it can still help the server recover its understanding of the key waveform morphology.

[0054] In some implementations, to avoid excessively large amounts of key summary data, edge nodes can further compress the key summary. For example, the key summary can be limited to include only a preset number of representative sampling points, such as 6, 8, or 10 representative sampling points; or the key summary can be limited to cover only a preset time window after a load mutation, such as the time range of 20us to 36us, excluding data from other stable intervals. Thus, the amount of data in the key summary can be significantly smaller than the complete original waveform data, while still retaining sufficient local morphological information to support further server-side determination.

[0055] In some implementations, after receiving the key summary, the server can use it in conjunction with previously received screening features. For example, the server can use the anomaly score in the screening features to determine that the current waveform has a high risk of anomaly, and then use the undershoot interval local sampling sequence and recovery interval local sampling sequence in the key summary to further analyze whether the current waveform has obvious overshoot, whether the recovery trend is smooth, whether abnormal ringing occurs, or whether it is necessary to request the complete original waveform data. Since the key summary provides supplementary information different from the screening features, the server can obtain more sufficient judgment criteria with lower data transmission overhead.

[0056] If the demand level is the second demand level, the screening features and associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

[0057] In a preferred embodiment, the server is configured to perform the following operations: Determine whether the identification of analog integrated circuits can be completed based on screening features. In a specific example, a large neural network model trained on a known method for identifying whether a circuit is normal can be deployed on the server first (e.g., a sequence classification model based on the Transformer architecture or a hybrid model combining a multilayer perceptron and an attention mechanism, hereinafter referred to as the identification model). The training process of the large model is roughly as follows: the server can collect a large number of samples from historical test records to build a training dataset for training the identification model.

[0058] Specifically, each training sample may contain at least the following two parts. The first part is the input features, which may include, but are not limited to: screening feature values ​​(e.g., peak value, settling time, anomaly score, voltage drop amplitude, overshoot amplitude), associated descriptive information (e.g., test type identifier, priority identifier), and optional normalized intermediate results. The second part is the annotation label, which can be provided by experienced test engineers or automated annotation systems after manually verifying historical waveforms. The annotation content may include the following two dimensions: first, the final judgment conclusion, i.e., whether the analog integrated circuit has an anomaly, such as "qualified", "questionable", or "unqualified"; second, the judgment confidence indicator, i.e., the assessment conclusion of the engineer or annotation system after reviewing the historical records on whether a reliable judgment can be made based solely on the screening features, such as "screening features are sufficient for judgment" or "screening features are insufficient for judgment, original waveform data is required".

[0059] After constructing the training dataset, the server can train the decision model in a supervised manner. Specifically, the training process can aim to minimize the cross-entropy loss between the model output and the labeled values, iteratively updating the model parameters through backpropagation. During training, data augmentation techniques can be used to expand the training samples, such as adding small random perturbations to the screening feature values ​​to improve the model's robustness to noise. Furthermore, regularization techniques such as Dropout can be used to reduce the risk of model overfitting.

[0060] In some implementations, the training dataset may include historical test samples corresponding to various test types (e.g., load step response test) and various anomaly levels, so that the decision model can cover a wide range of application scenarios. After the model is trained, the server can also use an independent validation dataset to evaluate the decision model to confirm that the model's classification accuracy, recall, and confidence calibration effect on unseen samples meet the preset requirements.

[0061] In a specific implementation, after receiving the screening features and associated description information sent by the edge nodes, the server can organize the information into the input format required by the decision model and input it into the trained decision model for inference. The decision model can output the following two types of results: first, a preliminary decision conclusion and its confidence probability for the analog integrated circuit; second, a judgment on whether the current screening features are sufficient to support a reliable decision, i.e., a "decision confidence flag". If the decision confidence flag is true, it means that the decision model believes that the current screening features are sufficient and the decision on the analog integrated circuit can be completed directly based on the screening features; if the decision confidence flag is false, it means that the screening features alone are not sufficient to make a reliable decision, and the server needs to further request key summaries or complete original waveform data.

[0062] In some implementations, the server can further set a confidence probability threshold, for example, setting the threshold to 0.90. When the confidence probability output by the judgment model is higher than 0.90, it can be considered that the current screening features are sufficient to support a reliable judgment, and the judgment confidence flag is set to true; when the confidence probability is lower than or equal to 0.90, the judgment confidence flag is set to false, and a subsequent data supplementation request process is triggered.

[0063] For example, in the first scenario, if the server receives the following screening characteristics for an analog integrated circuit: an abnormal peak value of 0.95V, a settling time of 24µs, an anomaly score of 1.00, and the test type is a load step response test, then the judgment model can analyze and conclude that the voltage drop of the circuit far exceeds the reference threshold, the settling time also exceeds the normal range, and the comprehensive anomaly score reaches full marks, indicating that multiple indicators are in a severely deviated state. In this case, the judgment model can output the conclusion of "judgment unqualified" with a high confidence probability (e.g., 0.97) and set the judgment confidence flag to true. This is because when multiple screening characteristics all point to severe anomalies, even without viewing the original waveform data, it is possible to determine with a high degree of confidence that the dynamic adjustment performance of the analog integrated circuit has obvious defects, and further obtaining the original waveform data (or a key summary of the original waveform data) will not change the above judgment.

[0064] For example, in the second scenario, if the server receives the following screening characteristics for an analog integrated circuit: an abnormal peak value of 1.05V, a setup time of 17µs, an anomaly score of 0.58, and the test type is a load step response test, then the judgment model can analyze and conclude that the overall anomaly score of the circuit is in the medium anomaly range. None of the screening characteristics reach a clearly acceptable level, nor a clearly unacceptable level, exhibiting a certain degree of ambiguity. In this case, the confidence probability output by the judgment model may be lower than the preset threshold (e.g., only 0.72), making it impossible to draw a definitive conclusion with sufficient certainty. This is because a moderate voltage drop and setup time delay may correspond to several different underlying causes: it could be normal performance of the circuit at a specific process corner, or it could be a slight anomaly caused by a deviation in compensation network parameters or insufficient loop gain margin, which is difficult to distinguish based solely on numerical characteristics such as peak value and setup time. Therefore, the judgment model can set the judgment confidence flag to false and trigger the server to further obtain a key summary of the original waveform data or request the original waveform data from the edge nodes to further analyze the ringing characteristics and overshoot patterns in the recovery interval, thereby assisting in making a more reliable judgment.

[0065] If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then continue to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

[0066] In a preferred embodiment, the server is also configured to perform the following operations: If a key summary of the received raw waveform data is determined, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the raw waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the raw waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the analog integrated circuit cannot be identified based on screening features and the key summary of the original waveform data, the remaining data of the original waveform data is requested from the edge node based on the associated description information. Those skilled in the art should understand that the remaining data of the original waveform data refers to the remaining original waveform data excluding the key summary of the original waveform data.

[0067] This invention provides an edge computing-based analog integrated circuit testing system, comprising edge nodes and a server, wherein the edge nodes are configured to perform the following operations: Obtain the step response waveform of an analog integrated circuit; Extract screening features from the step response waveform; Determine the abnormality level of the screening features and the type of test; The server's demand level for raw waveform data is determined based on the anomaly level of the screening features and the test type; and The strategy for sending screening features and raw waveform data to the server is determined based on the demand level.

[0068] In a preferred embodiment, the screening features include peak value, establishment time, and anomaly score; The edge nodes are also configured to perform the following operations: Generate associated descriptive information related to the screening features. The associated descriptive information includes the test object identifier, timestamp, and priority identifier.

[0069] In a preferred embodiment, the demand levels include a first demand level and a second demand level; Among them, the strategy for determining the transmission of screening features and raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, associated description information, and key summaries of the raw waveform data to the server. If the demand level is the second demand level, the screening features and associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

[0070] In a preferred embodiment, the server is configured to perform the following operations: Determine whether the identification of analog integrated circuits can be completed based on screening features; If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then continue to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

[0071] In a preferred embodiment, the server is also configured to perform the following operations: If a key summary of the received raw waveform data is determined, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the raw waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the raw waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the identification of analog integrated circuits cannot be completed based on screening features and key summaries of raw waveform data, the remaining data of the raw waveform data is requested from the edge node based on the associated description information.

[0072] To verify the effectiveness of the edge computing-based analog integrated circuit testing method proposed in this invention, a comparative experiment was conducted between the proposed solution and existing technical solutions based on a simulation platform.

[0073] In terms of the experimental environment, this invention uses Python 3.10 to build a data processing simulation environment for edge nodes and the server, and uses the NumPy and SciPy libraries to simulate the generation and feature extraction process of step response waveforms. The test waveform dataset was generated by LTspice XVII and contains 800 simulated load step response waveforms, covering three categories of samples: high anomaly, medium anomaly, and low anomaly. Specifically, there are 240 high anomaly samples, 320 medium anomaly samples, and 240 low anomaly samples. In the experiment, the communication channel bandwidth between the edge node and the server was set to a fixed value to allow for controlled comparison of data transmission volume and interaction rounds. The judgment model was trained based on 640 samples from the above dataset, and the remaining 160 samples were used for validation. Existing technical solutions use a method of "extracting features from the edge node and then uploading all original waveform data without discrimination" as a baseline.

[0074] Regarding the experimental results, statistical analysis shows that the present invention demonstrates considerable improvement over existing technologies in the following dimensions.

[0075] First, the average data transmission volume per transaction from the edge node to the server. In the solution of this invention, for samples with a demand level of second demand level (i.e., low-anomaly samples and some medium-anomaly samples), the edge node only sends screening features and associated description information to the server, without sending the original waveform data; for samples with a demand level of first demand level, the edge node only sends key summaries instead of the complete original waveform. Based on the statistics of all 800 test samples, the average data transmission volume per transaction of the solution of this invention is reduced by approximately 13.2% compared to the existing technology.

[0076] Second, the average waiting time required for the server to complete the judgment of a single sample. In existing technical solutions, the server needs to wait for the edge node to upload the complete original waveform before it can start the judgment process; however, in the solution of this invention, for samples whose confidence marker is true, the server can directly complete the judgment based on the screening features without waiting for the original waveform data to arrive. Based on the statistics of all test samples, the average waiting time of the server in the solution of this invention is reduced by approximately 11.8% compared with existing technical solutions.

[0077] Third, the number of cloud-edge interaction rounds. In existing technical solutions, when the screening features are insufficient to complete the judgment, the server needs to initiate a supplementary request, which increases the number of interaction rounds. However, in the solution of this invention, for samples with a demand level of first demand level, the edge node has already proactively attached the key summary when sending the first time, and the server does not need to initiate a supplementary request again in most cases. According to statistics of all test samples, the average number of cloud-edge interaction rounds of the solution of this invention is reduced by approximately 12.5% ​​compared with existing technical solutions.

[0078] Fourth, the overall test cycle time, which is the average time from the completion of waveform acquisition to the server outputting the final judgment. Considering both the reduced data transmission volume and the shortened server waiting time, the overall test cycle time of this invention is reduced by approximately 9.7% compared to existing technologies.

[0079] Fifth, accuracy. After introducing the judgment model and hierarchical data request mechanism, the server can still obtain key summaries and even complete original waveform data when needed. Therefore, the judgment accuracy does not significantly decrease due to the reduction in data transmission volume. On 160 verification samples, the judgment accuracy of this invention is 94.4%, which is basically the same as the 93.8% of existing solutions, and the difference is within the statistical error range.

[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and all such modifications or substitutions should be covered within the scope of protection of the present invention.

[0081] It should be noted that the functional modules involved in the various embodiments of the present invention can be integrated into the same processing unit, or each module can exist physically separately, or two or more modules can be integrated into the same unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0082] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0083] The present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for testing analog integrated circuits based on edge computing, the method being executed by edge nodes, the method comprising: Obtain the step response waveform of the analog integrated circuit; Extract screening features from the step response waveform; Determine the abnormality level and test type of the screening features; The server's demand level for raw waveform data is determined based on the anomaly level of the screening features and the test type; and The sending strategy for the screening features and the raw waveform data to be sent to the server is determined based on the demand level.

2. The method according to claim 1, wherein, The screening features include peak value, establishment time, and anomaly score; The method further includes: Generate association description information related to the screening features, including test object identifier, timestamp, and priority identifier.

3. The method according to claim 2, wherein, The demand levels include a first demand level and a second demand level; The strategy for determining the transmission of the screening features and the raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, the associated description information, and the key summary of the original waveform data to the server; If the demand level is the second demand level, then the screening features and the associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

4. The method according to claim 3, wherein, The server is configured to perform the following operations: Determine whether the analog integrated circuit can be determined based on the screening features; If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then it continues to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

5. The method according to claim 4, wherein, The server is also configured to perform the following operations: If a key summary of the original waveform data is received, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the original waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the original waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features and the key summary of the original waveform data, the remaining data of the original waveform data is requested from the edge node based on the associated description information.

6. An edge computing-based analog integrated circuit testing system, the system comprising edge nodes and a server, the edge nodes being configured to perform the following operations: Obtain the step response waveform of the analog integrated circuit; Extract screening features from the step response waveform; Determine the abnormality level and test type of the screening features; The server's demand level for raw waveform data is determined based on the anomaly level of the screening features and the test type; and The sending strategy for the screening features and the raw waveform data to be sent to the server is determined based on the demand level.

7. The system according to claim 6, wherein, The screening features include peak value, establishment time, and anomaly score; The edge node is also configured to perform the following operations: Generate association description information related to the screening features, including test object identifier, timestamp, and priority identifier.

8. The system according to claim 7, wherein, The demand levels include a first demand level and a second demand level; The strategy for determining the transmission of the screening features and the raw waveform data to the server based on the demand level includes: If the demand level is the first demand level, then send the screening features, the associated description information, and the key summary of the original waveform data to the server; If the demand level is the second demand level, then the screening features and the associated description information are sent to the server, and the original waveform data is saved locally on the edge node.

9. The system according to claim 8, wherein, The server is configured to perform the following operations: Determine whether the analog integrated circuit can be determined based on the screening features; If it is determined that the analog integrated circuit can be identified based on the screening features, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features, then it continues to determine whether the key summary of the original waveform data has been received; If it is determined that the key summary of the original waveform data has not been received, the original waveform data is requested from the edge node based on the associated description information.

10. The system according to claim 9, wherein, The server is also configured to perform the following operations: If a key summary of the original waveform data is received, then it is further determined whether the identification of the analog integrated circuit can be completed based on the screening features and the key summary of the original waveform data; If it is determined that the analog integrated circuit can be identified based on the screening features and the key summary of the original waveform data, then the identification of the analog integrated circuit is performed. If it is determined that the identification of the analog integrated circuit cannot be completed based on the screening features and the key summary of the original waveform data, the remaining data of the original waveform data is requested from the edge node based on the associated description information.

Images