Method and apparatus for lpddr grain power consumption test analysis based on energy efficiency model

Abstract

The embodiment of the application discloses a kind of LPDDR grain power consumption test analysis method and equipment based on energy efficiency model, method includes: the original power consumption response data sequence of the time stamp being produced when being tested LPDDR grain operation test incentive program is collected;It is mapped to three-dimensional space with time as horizontal axis, power consumption as vertical axis, power consumption rate of change as longitudinal axis, generates power consumption response space-time mapping surface structure;Call LPDDR grain energy efficiency model to feature analysis of surface, extract transient power consumption impact response feature sequence and steady-state power consumption response feature sequence;Two feature sequences are matched in standard energy efficiency state response template library, and energy efficiency state transition path descriptor is generated;According to the path descriptor, the surface is segmented, and the power consumption distribution characteristic parameter set of each node position is obtained, and the power consumption test analysis conclusion mark containing energy efficiency level mark is generated based on the set, the fine identification and automatic quantitative rating of LPDDR grain energy efficiency state are realized.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a method and device for testing and analyzing the power consumption of LPDDR chips based on an energy efficiency model. Background Technology

[0002] With the continuous growth in demand for memory bandwidth and capacity from mobile smart terminals, high-performance computing, and data centers, Low Power Double Data Rate (LPDDR) chips have been widely adopted due to their excellent balance between power consumption and performance. As an important type of Dynamic Random Access Memory (DRAM), the power consumption characteristics of LPDDR chips directly affect the battery life and thermal design of terminal devices. Therefore, accurate power consumption testing and analysis of LPDDR chips has become an indispensable part of memory chip design verification, production screening, and system-level power optimization. Existing LPDDR chip power consumption testing techniques typically include power consumption measurement based on average current, transient response observation based on time-domain waveform analysis, and power consumption simulation based on standard test vectors.

[0003] However, the aforementioned existing technologies are difficult to systematically correlate the raw power consumption response data of LPDDR chips with their internal energy efficiency state transition paths, and thus lack an effective method for automatically evaluating the overall energy efficiency level of the chips. Summary of the Invention

[0004] This application provides a method and device for testing and analyzing the power consumption of LPDDR chips based on an energy efficiency model.

[0005] This application provides, in one aspect, a method for testing and analyzing the power consumption of LPDDR chips based on an energy efficiency model, applied to computer equipment. The method includes:

[0006] The raw power consumption response data sequence is collected when the LPDDR chip under test runs the test stimulus program under preset test conditions. The raw power consumption response data sequence includes multiple power consumption sampling point values ​​with timestamps collected continuously.

[0007] The original power consumption response data sequence is mapped to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and the rate of change of power consumption value as the depth axis, to generate the spatiotemporal mapping surface structure of the power consumption response of the LPDDR chip under test under the test conditions.

[0008] The pre-built LPDDR particle energy efficiency model is invoked to perform surface feature analysis on the power consumption response spatiotemporal mapping surface structure, and the transient power consumption impulse response feature sequence and steady-state power consumption response feature sequence are extracted from the power consumption response spatiotemporal mapping surface structure.

[0009] Based on the transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence, template matching processing is performed in the standard energy efficiency state response template library of the LPDDR particle energy efficiency model to generate an energy efficiency state transition path descriptor for the LPDDR particle under test to switch from the current energy efficiency state to the target energy efficiency state.

[0010] The power consumption response spatiotemporal mapping surface structure is segmented according to the energy efficiency state transition path descriptor to obtain a set of power consumption distribution characteristic parameters of each node position of the LPDDR chip under test on the energy efficiency state transition path, and a power consumption test analysis conclusion identifier containing the energy efficiency level identifier is generated based on the set of power consumption distribution characteristic parameters.

[0011] One embodiment of this application provides a computer device, including:

[0012] A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement any of the described energy efficiency model-based LPDDR chip power consumption test and analysis methods.

[0013] One embodiment of this application provides a readable storage medium on which a program or instruction is stored. When the program or instruction is executed by a processor, it implements the steps of the LPDDR chip power consumption test and analysis method based on the energy efficiency model. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating a power consumption test and analysis method for LPDDR chips based on an energy efficiency model, provided in an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of the basic structure of a computer device provided in an embodiment of this application.

[0017] Figure 3 This is a functional block diagram of an LPDDR chip power consumption test and analysis device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] Please see Figure 1 , Figure 1 This is a flowchart of an LPDDR chip power consumption test and analysis method based on an energy efficiency model provided in an embodiment of this application. The method can be executed by a computer device or by a computer device and a server. The method can include steps 110-160.

[0020] Step 110: Collect the raw power consumption response data sequence generated when the LPDDR chip under test runs the test stimulus program under preset test conditions. The raw power consumption response data sequence includes multiple power consumption sampling point values ​​with timestamps collected continuously.

[0021] In this embodiment of the application, the computer device first needs to construct a standardized test environment. The core of this environment is a predefined test condition. This predefined test condition is not a single fixed condition, but a multi-dimensional set of parameters, including but not limited to the set ambient temperature value, the power supply voltage value of the chip, and a test stimulus program consisting of a series of set command sequences issued by the memory controller. This test stimulus program is designed to simulate the typical workload of LPDDR chips in actual applications, such as continuous burst write operations, random read operations, periodic automatic refresh operations, and sequences of waking up from deep sleep mode.

[0022] When the LPDDR chip under test begins executing the test stimulus program under this preset test condition, its internal circuits, such as the memory array, address decoder, and input / output buffers, frequently switch between active and sleep states, causing dynamic changes in the instantaneous power consumption on the power supply pins. A computer device is connected to the chip's power path via a high-time-resolution data acquisition card, synchronously acquiring the real-time current and voltage flowing through the chip at a preset sampling frequency, and instantaneously calculating the corresponding power consumption value. Each acquisition yields a power consumption value represented by a double-precision floating-point number, and simultaneously records a nanosecond-level timestamp provided by a high-precision clock source corresponding to that sampling moment.

[0023] After a complete test cycle, such as the entire time window from the start to the end of the test stimulus program, the computer device obtains an ordered list of data, namely the raw power response data sequence. In this sequence, each element is a data pair containing a timestamp and a power consumption value. For example, the first data pair can be represented as a combination of T1 and P1, the second as a combination of T2 and P2, and so on, until the Nth data pair is represented as a combination of Tn and Pn, where T represents the timestamp in seconds and P represents the power consumption value in milliwatts. This sequence continuously and completely depicts the raw trajectory of the instantaneous power consumption of the particle under test under target operating conditions as it evolves over time.

[0024] Step 120: Map the original power consumption response data sequence to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and the rate of change of power consumption value as the depth axis, to generate the spatiotemporal mapping surface structure of the power consumption response of the LPDDR chip under test under the test conditions.

[0025] After acquiring the raw power response data sequence, the computer device does not directly analyze the original time-numerical sequence. Instead, it performs a crucial data space transformation, elevating the one-dimensional time sequence to three-dimensional space and converting it into a surface structure with geometric topological information. This process is achieved through the following steps:

[0026] Step 121: Perform time axis coordinate allocation processing on the power consumption sampling point value in the original power consumption response data sequence, convert the timestamp mark corresponding to each power consumption sampling point value into the time coordinate component value in the three-dimensional space, and obtain the time coordinate positioning point corresponding to each power consumption sampling point value.

[0027] For each data pair in the original power response data sequence, such as any data pair Ti and Pi, the computer device first extracts its timestamp Ti. To locate this point in a three-dimensional Cartesian coordinate system, Ti is directly mapped to the coordinate components of the point on a predefined time axis, typically defined as the X-axis, with units of seconds. Therefore, the time coordinate location corresponding to this data point is Xi equal to Ti. This process essentially assigns an absolute spatial position in the time dimension to each power consumption sampling point, solidifying the originally linearly flowing time into a measurable spatial dimension in three-dimensional space.

[0028] Step 122: Perform power consumption value axis coordinate allocation processing on the power consumption sampling point value in the original power consumption response data sequence, and use the power consumption sampling point value as the power consumption value coordinate component value in the three-dimensional space to obtain the power consumption value coordinate positioning point corresponding to each power consumption sampling point value.

[0029] For the same data pair Ti and Pi, the computer device extracts its power consumption value Pi, which is directly mapped to the coordinate components of the midpoint in three-dimensional space on the power consumption value axis. The power consumption value axis is usually defined as the Y-axis, with units of milliwatts. Thus, the power consumption value coordinate location point corresponding to this data point is Yi equal to Pi. This step assigns a spatial position to each sampling point in the power consumption amplitude dimension, so that the magnitude of the power consumption value is directly reflected in the height of the midpoint in space.

[0030] Step 123: Perform change rate calculation processing on the values ​​of adjacent power consumption sampling points in the original power consumption response data sequence. Calculate the difference between each power consumption sampling point value and its previous power consumption sampling point value, and divide the difference by the interval duration of the timestamps corresponding to the adjacent power consumption sampling point values ​​to obtain the power consumption value change rate corresponding to each power consumption sampling point value. Use the power consumption value change rate as the change rate coordinate component value in the three-dimensional space to obtain the change rate coordinate positioning point corresponding to each power consumption sampling point value.

[0031] This step constructs the third dimension of the three-dimensional space and is crucial for reflecting the dynamic characteristics of power consumption. The computer device starts from the second data point in the sequence and traverses the entire original power response data sequence. For the currently traversed i-th data point, where i ranges from 2 to n, the computer device performs the following calculations: First, it calculates the difference between the power consumption value Pi at this point and the power consumption value P(i-1) at the previous point, obtaining the power consumption change, denoted as DeltaPi equal to Pi minus P(i-1). Then, it calculates the difference in timestamps between these two points, obtaining the time interval, denoted as DeltaTi equal to Ti minus T(i-1). Finally, it divides the power consumption change by the time interval to obtain the rate of change of power consumption at that point, denoted as Ri equal to DeltaPi divided by DeltaTi. The physical meaning of the rate of change Ri is the average rate of change of power consumption per unit time, with the dimension milliwatts per second.

[0032] For the first point in the sequence, i.e., when i equals 1, since it has no predecessor point, its rate of change R1 can be set to a preset initial value, such as zero. The calculated Ri is then mapped to the coordinate component Zi of that point on the rate of change axis, which is equal to Ri. The rate of change axis is usually defined as the Z-axis, with units of milliwatts per second. Thus, each of the original data points Ti and Pi has been assigned three coordinate components, namely Xi, Yi, and Zi, thereby completing the transformation from one-dimensional time series data to three-dimensional spatial point data.

[0033] Step 124: Perform three-dimensional spatial point construction processing on the time coordinate positioning point, the power value coordinate positioning point and the rate of change coordinate positioning point corresponding to each power consumption sampling point value to generate a spatial coordinate point representation of each power consumption sampling point value in the three-dimensional space.

[0034] The computer device integrates the three coordinate components calculated for the same original data point in steps 121 to 123 to form a complete three-dimensional spatial coordinate point. For example, for the i-th data point, its spatial coordinate point is represented as Point_i, which is an ordered combination of three coordinate values, specifically Xi, Yi, and Zi. In this way, the entire original power response data sequence is converted into a discrete set of points in three-dimensional space, represented as the set Points, which contains all spatial points from Point_1 to Point_n.

[0035] Step 125: Perform sequential connection processing on the spatial coordinate point representations corresponding to all power consumption sampling point values ​​in the original power consumption response data sequence. Connect adjacent spatial coordinate point representations with straight line segments according to the order of time coordinate positioning points to obtain the power consumption response spatial trajectory line of the original power consumption response data sequence in the three-dimensional space.

[0036] After obtaining the discrete point set Points, the computer device connects adjacent points in space according to the chronological order of the time coordinates, that is, according to the order of the size of the Xi coordinate values. Specifically, Point_1 and Point_2 are connected by a straight line segment, then Point_2 and Point_3 are connected, and so on, until Point_(n-1) and Point_n are connected. Thus, the implicit time order in the original sequence is represented as a continuous spatial polyline in three-dimensional space, consisting of straight line segments connected end to end. This polyline is the power consumption response spatial trajectory line, which intuitively shows the evolution path of the power consumption state of the LPDDR chip in three-dimensional space, where each turning point corresponds to an original power consumption sampling point.

[0037] Step 126: Perform surface expansion and smoothing processing based on the power response spatial trajectory line to construct a power response spatiotemporal mapping surface structure.

[0038] After obtaining the power response spatial trajectory, the computer device performs further geometric processing to generate a continuous surface. One implementation is to use this trajectory as the core ridge of the surface, and then expand around this ridge in a plane perpendicular to its tangent direction according to predetermined rules. The expansion rules are determined as follows: First, calculate the tangent direction vector at each point on the trajectory. Then, construct a plane passing through the point and perpendicular to the tangent. Within this plane, define a circular region centered at the point. The radius of this circular region is a variable parameter, its value related to the rate of change of power consumption at that point. Specifically, a larger rate of change means a more drastic power consumption jump, and the more the local geometric features of that point should be emphasized; therefore, the radius should be smaller, making the subsequently constructed surface more closely adhere to the trajectory near that point, forming a sharp "ridge." Conversely, a smaller rate of change means a more stable power consumption, and the radius should be larger, making the surface expand more gently near that point. After determining the expansion radius at each point, the computer device generates a circular cross-section in each of these planes. Then, using a sweeping algorithm, the edges of these sequentially arranged circular cross-sections are smoothly connected to form a continuous tubular surface.

[0039] Another approach is to use the trajectory line and its projection line on the time-power plane as two boundary curves. The computer device first projects each point of the trajectory line vertically onto the XY plane, obtaining a series of projection points. Connecting these projection points sequentially yields the projection line. Then, between the trajectory line and the projection line, a smooth surface patch is constructed using bilinear interpolation or a B-spline surface fitting algorithm. Specifically, the trajectory line and the projection line serve as two opposing boundaries of the surface. Multiple isoparametric lines are uniformly inserted between them, and these isoparametric lines and the boundary lines together define the surface's mesh. Finally, these meshes are smoothed so that the surface can encompass all points traversed by the original trajectory line and smoothly fill the gaps between points. After this expansion and smoothing process, a continuous and smooth power response spatiotemporal mapping surface structure is obtained. The projection of this surface onto the time-power plane reflects the macroscopic trend of power consumption, while its depth-direction changes contain information about the dynamic rate of change of power consumption, providing rich geometric data for feature extraction.

[0040] Step 130: Call the pre-built LPDDR particle energy efficiency model to perform surface feature analysis processing on the power consumption response spatiotemporal mapping surface structure, and extract the transient power consumption impact response feature sequence and steady-state power consumption response feature sequence from the power consumption response spatiotemporal mapping surface structure.

[0041] After constructing the power consumption response spatiotemporal mapping surface structure, the computer device calls a pre-trained LPDDR particle energy efficiency model to perform in-depth analysis of this surface. This energy efficiency model itself encapsulates a series of algorithms for identifying and extracting key power consumption features. The specific analysis process includes the following steps:

[0042] Step 131: Perform surface curvature calculation processing on the power consumption response spatiotemporal mapping surface structure, calculate the Gaussian curvature value and average curvature value of each surface grid vertex on the power consumption response spatiotemporal mapping surface structure, and obtain the surface curvature distribution field composed of the Gaussian curvature value and average curvature value of all surface grid vertices.

[0043] First, the computer device discretizes the continuous power response spatiotemporal mapping surface structure into a mesh surface composed of numerous tiny triangular grids. This discretization process is achieved by selecting sampling points on the surface at a certain resolution, and then connecting these sampling points into a triangular mesh according to certain topological rules. For each mesh vertex, the computer device uses the three-dimensional spatial coordinate information of all its neighboring vertices to calculate the Gaussian curvature and mean curvature value of that point using standard algorithms in differential geometry. In the specific calculation process, for any vertex, it first needs to obtain its directly adjacent ring of vertices, and then use the coordinates of these neighboring vertices and the vertex's own coordinates to construct a locally parameterized plane.

[0044] Based on this, the Gaussian curvature at the vertex is calculated by solving for the coefficients of the first and second fundamental forms. This curvature is equal to the product of the principal curvatures, reflecting the overall degree of curvature of the surface at that point. Simultaneously, the mean curvature is calculated, which is equal to the average of the principal curvatures, reflecting the average curvature of the surface at that point in all directions. After performing this calculation on all mesh vertices, each point on the entire surface is assigned both Gaussian and mean curvature values, thus forming a curvature distribution field covering the entire surface. This curvature distribution field is a scalar field, where each point is associated with two values.

[0045] Step 132: Perform extreme point detection processing on the surface curvature distribution field, identify the surface grid vertices in the surface curvature distribution field whose Gaussian curvature value or average curvature value is greater than a preset curvature threshold as candidate curvature extreme points, and record the three-dimensional spatial coordinate position of each candidate curvature extreme point on the power consumption response spatiotemporal mapping surface structure.

[0046] After obtaining the curvature distribution field, the computer device begins searching for regions with abnormally high curvature values. These regions typically correspond to drastic power consumption jumps, i.e., transient impact events. To this end, the device sets a curvature threshold, which is not a static, fixed value but dynamically determined based on the statistical characteristics of the entire curvature distribution field. For example, the mean and standard deviation of the Gaussian curvature values ​​of all vertices on the entire surface can be calculated, and the threshold can then be set as the mean plus a certain multiple of the standard deviation, or as a high quantile of all curvature values. A similar approach is used to set the threshold for the average curvature. The computer device traverses all grid vertices. If a vertex's Gaussian curvature value is greater than the set Gaussian curvature threshold, or its average curvature value is greater than the set average curvature threshold, then the vertex is marked as a candidate curvature extremum point, and its precise coordinates in three-dimensional space, i.e., its X, Y, and Z coordinate components, are recorded.

[0047] Step 133: Perform spatial clustering on the candidate curvature extrema points, grouping candidate curvature extrema points with adjacent three-dimensional spatial coordinates into a curvature extrema point cluster. Each curvature extrema point cluster corresponds to a transient power consumption impact event region. Perform cluster center coordinate calculation on each curvature extrema point cluster, and use the arithmetic mean of the three-dimensional spatial coordinates of all candidate curvature extrema points in each curvature extrema point cluster as the cluster center coordinates of the curvature extrema point cluster, and use the cluster center coordinates as the transient power consumption impact response feature points.

[0048] Because a transient event, such as a transition from sleep to wake-up, typically forms a continuous high-curvature region on a surface rather than an isolated point, the candidate extrema detected in the previous step often appear in clusters. The computer device uses a spatial clustering algorithm to cluster these discrete candidate extrema. The specific clustering process is as follows: First, arbitrarily select a point from the candidate extrema set as the starting point, and then set a distance threshold centered on that point. Search for all other candidate points whose spatial Euclidean distance to that point is less than the threshold, and group these points into the same cluster. Then, using the newly added cluster as the center, continue searching outwards until no new points can be added to the cluster, thus forming a cluster.

[0049] Then, the above process is repeated for the remaining points until all points are assigned to a cluster. Each cluster represents the surface region where an independent transient power surge event occurred. For each cluster, the computer device calculates its geometric center, i.e., the cluster center coordinates. Specifically, the arithmetic mean of the X coordinates of all candidate extrema points within the cluster is obtained by summing the X coordinates and dividing by the total number of points in the cluster, and this is taken as the X coordinate of the cluster center. Similarly, the same arithmetic mean is calculated for the Y and Z coordinates to obtain the Y and Z coordinates of the cluster center, respectively. The calculated center point coordinates are defined as a transient power surge response feature point, representing the core location of the transient surge event in three-dimensional space.

[0050] Step 134: Sort the cluster center coordinates of all curvature extreme point clusters on the power consumption response spatiotemporal mapping surface structure according to the order of time coordinate positioning points to generate a transient power consumption impulse response feature sequence composed of multiple transient power consumption impulse response feature points arranged in time order.

[0051] After acquiring all transient power consumption surge response characteristic points, the computer device sorts these points in ascending order based on their X-coordinate, i.e., the time coordinate component. The resulting ordered list is the transient power consumption surge response characteristic sequence. Each element in this sequence is a three-dimensional coordinate point, abstractly representing the temporal sequence and spatial location of all drastic power consumption jump events occurring in the LPDDR chip under test within the test time window.

[0052] Step 135: Perform flat region detection and projection processing on the power consumption response spatiotemporal mapping surface structure to extract the steady-state power consumption response feature sequence.

[0053] In parallel or sequentially with the extraction of transient features, the computer device also extracts another type of feature from the surface. Specifically, step 135 is implemented through the following sub-steps:

[0054] Step 1351: Perform flat region detection processing on the power consumption response spatiotemporal mapping surface structure, and identify continuous surface regions in the power consumption response spatiotemporal mapping surface structure where both the Gaussian curvature value and the average curvature value are less than the preset flat curvature threshold as candidate flat regions.

[0055] The computer device sets a flat curvature threshold, which is also dynamically determined based on the statistical characteristics of the curvature distribution field, but is usually much smaller than the extreme point detection threshold used in step 132. For example, the median of the absolute values ​​of the Gaussian curvature values ​​of all vertices on the entire surface can be calculated, and a multiple of this median can be used as the Gaussian curvature flatness threshold. A similar method is used to set the flatness threshold for the average curvature. The device then traverses all mesh vertices again. If the absolute value of a vertex's Gaussian curvature value is less than the set Gaussian curvature flatness threshold, and the absolute value of its average curvature value is also less than the set average curvature flatness threshold, then the vertex is considered to be in a locally flat region. All vertices satisfying this condition are initially marked.

[0056] Step 1352: Perform regional connectivity analysis on the candidate flat regions, merge spatially connected candidate flat regions into the same flat region unit, and record the three-dimensional spatial coordinate range of each flat region unit on the power consumption response spatiotemporal mapping surface structure.

[0057] The computer device performs connectivity analysis on all the flat vertices marked in the previous step. The connectivity analysis uses a region growing algorithm: starting from a flat vertex, it checks its neighboring vertices; if a neighboring vertex is also marked as a flat vertex, it merges them into the same connected component. Then, based on the newly merged vertices, it continues checking outwards until no neighboring flat vertices are found. Thus, a connected component constitutes a flat region unit. This process is repeated until all flat vertices are assigned to a flat region unit. For each flat region unit, the device records the 3D spatial coordinate range covered by its boundary, i.e., the minimum and maximum X-coordinates, Y-coordinates, and Z-coordinates of all vertices within the unit. This coordinate range precisely defines the position and size of the flat region in 3D space.

[0058] Step 1353: Perform coordinate projection processing on the three-dimensional spatial coordinate range corresponding to each flat region unit, projecting the three-dimensional spatial coordinate range of each flat region unit onto a two-dimensional plane composed of the time axis and the power consumption value axis, to obtain the start time coordinate value and end time coordinate value of each flat region unit on the time axis, and the start power consumption coordinate value and end power consumption coordinate value on the power consumption value axis.

[0059] Each flat region can be understood as an irregular curved surface in three-dimensional space. To quantify its characteristics in terms of steady-state power consumption, the computer device projects it onto the XY plane, i.e., the time-power plane. The projection operation ignores the Z-coordinate of each point, considering only its X and Y coordinates. The result of the projection is the projection interval of the region on the X-axis, with its starting coordinate being the minimum X-coordinate and its ending coordinate being the maximum X-coordinate. Similarly, on the Y-axis, the starting coordinate is the minimum Y-coordinate and its ending coordinate is the maximum Y-coordinate. The minimum and maximum X-coordinates represent the start and end times of this steady-state phase, while the minimum and maximum Y-coordinates represent the range of power consumption fluctuations within this steady-state phase.

[0060] Step 1354: Generate the steady-state power response feature interval corresponding to each flat region unit based on the start time coordinate value, end time coordinate value, start power consumption coordinate value, and end power consumption coordinate value of each flat region unit, and arrange the steady-state power response feature intervals corresponding to all flat region units in chronological order to generate a steady-state power response feature sequence.

[0061] The computer device formalizes the projection result of each flat region cell as a feature interval, which is a quadruple containing the start time coordinate, end time coordinate, start power consumption coordinate, and end power consumption coordinate. Then, based on the start time coordinate of each interval, these feature intervals are sorted in chronological order. The resulting ordered list is the steady-state power consumption response feature sequence. Each element in this sequence is a quadruple, abstractly representing the various stages of relatively stable power consumption of the LPDDR chip under test during the test, along with their duration and amplitude range.

[0062] Step 140: Based on the transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence, perform template matching processing in the standard energy efficiency state response template library of the LPDDR particle energy efficiency model to generate an energy efficiency state transition path descriptor for the LPDDR particle under test to switch from the current energy efficiency state to the target energy efficiency state.

[0063] After obtaining the transient and steady-state characteristic sequences of the particles under test, the computer equipment compares them with a built-in standard template to identify the transition process of their energy efficiency state. This is achieved through the following steps:

[0064] Step 141: Convert the three-dimensional spatial coordinate position of each transient power consumption impulse response feature point in the transient power consumption impulse response feature sequence into a transient feature vector with time coordinate components, power consumption numerical coordinate components and rate of change coordinate components as elements, to obtain a transient feature vector sequence composed of multiple transient feature vectors.

[0065] For each feature point in the transient feature sequence, the feature point itself is a three-dimensional spatial coordinate point, containing X, Y, and Z coordinates. The computer device directly formalizes it as a three-dimensional feature vector. The first component of this vector is the time coordinate, the second component is the power consumption value coordinate, and the third component is the rate of change coordinate. Thus, the entire transient feature sequence is transformed into a sequence composed of feature vectors, the length of which is the same as the number of transient feature points.

[0066] Step 142: Convert each steady-state power response feature interval in the steady-state power response feature sequence into a steady-state feature vector with the start time coordinate, end time coordinate, start power coordinate, and end power coordinate as elements, to obtain a steady-state feature vector sequence composed of multiple steady-state feature vectors.

[0067] For each feature interval in the steady-state feature sequence, this interval is a quadruple containing the start time, end time, start power consumption, and end power consumption. The computer device directly formalizes this as a four-dimensional feature vector, where the first component is the start time coordinate, the second component is the end time coordinate, the third component is the start power consumption coordinate, and the fourth component is the end power consumption coordinate. Thus, the entire steady-state feature sequence is transformed into a steady-state feature vector sequence composed of multiple four-dimensional feature vectors.

[0068] Step 143: Call the standard energy efficiency state response template library stored in the LPDDR particle energy efficiency model. The standard energy efficiency state response template library contains multiple standard energy efficiency state response templates. Each standard energy efficiency state response template corresponds to a standard energy efficiency state type and contains a standard transient feature vector sequence and a standard steady-state feature vector sequence.

[0069] The computer device accesses a database pre-existing in the LPDDR chip energy efficiency model, namely the standard energy efficiency state response template library. This library stores various standard energy efficiency state response templates summarized from extensive standard testing and calibration for the same type of LPDDR chips. For example, there is a template representing the "idle state," which may contain a relatively long sequence of standard steady-state feature vectors with low power consumption, while the sequence of standard transient feature vectors is very short or even empty. Another example is a template representing the "active read state," which contains a standard transient feature vector followed by a standard steady-state feature vector, where the transient feature has a large Z-coordinate (rate of change), while the steady-state feature has a moderate power consumption. Each template is associated with its corresponding standard energy efficiency state type identifier.

[0070] Step 144: Perform sequence similarity matching processing on the transient feature vector sequence and the standard transient feature vector sequence of each standard energy efficiency state response template in the standard energy efficiency state response template library, calculate the DTW distance between the transient feature vector sequence and each standard transient feature vector sequence, and generate a transient matching similarity based on the DTW distance to obtain a transient matching similarity set composed of multiple transient matching similarities.

[0071] The computer device compares the transient feature vector sequence to be tested with the standard transient feature vector sequence of each template in the template library. Since the lengths of the two sequences may differ, and the timing of the transient impact may have slight shifts, the device uses a dynamic time warping (DTW) algorithm to calculate the similarity between the two sequences. The core of the DTW algorithm is to construct a distance matrix, where the rows correspond to the vectors of the test sequence, and the columns correspond to the vectors of the standard sequence. The value of each element in the matrix is ​​the Euclidean distance between the i-th vector in the test sequence and the j-th vector in the standard sequence. Then, using dynamic programming, a path with the minimum cumulative distance is found in the matrix from the starting point (1,1) to the ending point (M, M_std). This path represents the optimal non-linear alignment between the two sequences. The sum of the distances of all points on the path is the DTW distance. The smaller this DTW distance, the more similar the shapes of the two sequences are. Subsequently, the device converts this distance into a transient matching similarity using an inverse proportional function. This function ensures that the similarity is at its maximum when the distance is zero, and approaches zero when the distance approaches infinity. Perform this operation on all templates in the template library to obtain a transient matching similarity set, where each element in the set corresponds to a transient matching similarity of a template.

[0072] Step 145: Perform interval overlap matching processing on the steady-state feature vector sequence and the standard steady-state feature vector sequence of each standard energy efficiency state response template in the standard energy efficiency state response template library. Calculate the interval overlap area ratio between each steady-state feature vector in the steady-state feature vector sequence and the corresponding standard steady-state feature vector in each standard steady-state feature vector sequence. Generate a steady-state matching similarity based on the interval overlap area ratio to obtain a steady-state matching similarity set composed of multiple steady-state matching similarities.

[0073] In parallel with transient matching, the computer device performs steady-state feature matching. For a steady-state feature vector sequence to be tested and a steady-state feature vector sequence of a template in a template library, the device first needs to map the steady-state stages of the two sequences. An exemplary mapping method is to assume that the two sequences have the same number of stages and are sequentially corresponding. If the number is different, the optimal stage mapping relationship can be found using dynamic programming. For each pair of corresponding stages, the degree of overlap between them on the time-power plane is calculated. Specifically, for the k-th stage to be tested, its start time and end time, as well as its start power and end power, are known. For the corresponding template stage, its corresponding standard start time and end time, as well as its standard start power and standard end power, are also known. The overlap time is calculated as the difference between the larger of the two start times and the smaller of the two end times. If this difference is positive, there is time overlap; otherwise, there is no overlap. Similarly, the overlap power is calculated as the difference between the larger of the two start power consumptions and the smaller of the two end power consumptions. Multiplying the time overlap length by the power overlap length yields the overlap area. The area of ​​a template stage is obtained by multiplying the standard time length by the standard power consumption range. The overlap of a single stage is the overlap area divided by the area of ​​the template stage. The steady-state matching similarity of the entire sequence can be taken as the weighted average of the overlaps of all stages, with the weights being the duration of each stage, i.e., the standard time length. Performing this operation on all templates in the template library yields a steady-state matching similarity set.

[0074] Step 146: Perform weighted fusion processing on the transient matching similarity set and the steady-state matching similarity set, assign a first fusion weight coefficient to each transient matching similarity, assign a second fusion weight coefficient to each steady-state matching similarity, and calculate the comprehensive matching similarity corresponding to each standard energy efficiency state response template.

[0075] For each template in the template library, the computer device fuses its corresponding transient and steady-state matching similarities. During fusion, different weight coefficients are assigned to the transient and steady-state components. The weight assignment can be based on prior knowledge; for example, transient features are more critical for identifying energy efficiency state transitions and therefore can be given higher weights. The sum of the two weight coefficients must equal one. The overall matching similarity equals the transient matching similarity multiplied by the transient weight plus the steady-state matching similarity multiplied by the steady-state weight. After performing this weighted fusion calculation on all templates, a set of overall matching similarities is obtained.

[0076] Step 147: Based on the comprehensive matching similarity, select the standard energy efficiency state response template with the highest comprehensive matching similarity from the standard energy efficiency state response template library as the target matching template, and extract the standard energy efficiency state type identifier corresponding to the target matching template.

[0077] The computer device identifies the maximum value in the comprehensive matching similarity set, and the template corresponding to this maximum value is determined as the target matching template. The device then reads the standard energy efficiency state type identifier associated with this template from the template library. This identifier is a code used to uniquely distinguish different energy efficiency state types.

[0078] Step 148: Based on the standard energy efficiency state type identifier corresponding to the target matching template and the standard energy efficiency state transition relationship recorded in the target matching template, generate an energy efficiency state transition path descriptor for the LPDDR chip under test to switch from the current energy efficiency state to the target energy efficiency state.

[0079] The standard energy efficiency state response template not only includes the characteristics of a single state but also implicitly or explicitly records the standard transition relationships between energy efficiency states. For example, a template might record the complete process from an "idle state" through a transient state to an "active read state." The computer device matches this series of state transitions recorded in the target matching template with the template's own type identifier to generate a structured descriptor. This descriptor can be a string containing the type identifiers of each energy efficiency state experienced in chronological order, and uses defined connectors to indicate the direction of transition between states. This descriptor accurately summarizes the energy efficiency state transition path exhibited by the particle under test in this test.

[0080] Step 150: Perform surface region segmentation processing on the power consumption response spatiotemporal mapping surface structure according to the energy efficiency state transition path descriptor to obtain the power consumption distribution characteristic parameter set of the LPDDR chip under test at each node position on the energy efficiency state transition path.

[0081] After determining the energy efficiency state transition path, the computer device returns the original power consumption response spatiotemporal mapping surface, and performs refined region segmentation and parameter extraction along this path. The specific process is as follows:

[0082] Step 151: Parse the multiple energy efficiency state transition node identifiers contained in the energy efficiency state transition path descriptor. Each energy efficiency state transition node identifier corresponds to a key location point on the energy efficiency state transition path.

[0083] The computer device first parses the energy efficiency state transition path descriptor generated in step 148. For example, if the descriptor is in the format of "state identifier A-state identifier B-state identifier C", then three key node identifiers can be parsed from this descriptor: the starting node identifier A, the intermediate node identifier B, and the ending node identifier C. Each node identifier corresponds to an energy efficiency state on the path.

[0084] Step 152: Based on the energy efficiency state transition node identifier, query the three-dimensional spatial coordinate position corresponding to the energy efficiency state transition node identifier from the power consumption response spatiotemporal mapping surface structure to obtain the transition node coordinates of each energy efficiency state transition node.

[0085] After parsing the node identifiers, the computer device needs to map these abstract identifiers back to specific locations on the surface. This mapping process depends on the template matching results in step 140. In step 140, when the target matching template is determined, the alignment relationship between the transient and steady-state feature sequences to be tested and the standard sequences in the template is also determined. Through this alignment relationship, it is possible to trace back to the specific region on the power response spatiotemporal mapping surface corresponding to each node identifier. For example, for the node identifier "state identifier A", the matching steady-state feature interval to be tested can be found based on the standard steady-state feature interval corresponding to this state in the template. Then, the center point of this interval on the surface, or the arithmetic mean of the coordinates of all points within the interval, is taken as the transition node coordinates of node identifier A. These coordinates are also a three-dimensional spatial point containing X, Y, and Z components.

[0086] Step 153: Using the coordinates of the transition node of each energy efficiency state transition node as the center, delineate the surface region segmentation range centered on the coordinates of the transition node on the power consumption response spatiotemporal mapping surface structure, and obtain the surface segmentation sub-region corresponding to each energy efficiency state transition node.

[0087] After obtaining the node coordinates, the computer device delineates a neighborhood on the surface centered on each node as a segmented sub-region. The method for delineating this range is based on geodesic distances on the surface. First, the geodesic distances of all other points on the surface relative to the center point need to be calculated. Geodesic distance refers to the length of the shortest path along the surface. Then, a distance threshold is defined. This threshold is not a fixed value but is dynamically adjusted based on the local geometric features of the surface at the node. Specifically, the absolute value of the Gaussian curvature at the node can be calculated. The larger the absolute value of the curvature, the more drastic the surface change at that point, and the more "concentrated" the state represented by the node. Therefore, the distance threshold should be smaller to ensure that the segmented sub-region focuses on the core area of ​​that state. Conversely, the smaller the absolute value of the curvature, the larger the threshold. All points whose geodesic distance from the center point is less than this threshold collectively form a surface patch, which is the segmented sub-region corresponding to that node.

[0088] Step 154: Perform region boundary identification processing on the surface segmented sub-region corresponding to each energy efficiency state transition node, extract the boundary contour line of each surface segmented sub-region, and determine the coverage range of each surface segmented sub-region on the power consumption response spatiotemporal mapping surface structure based on the boundary contour line.

[0089] For each surface patch defined in the previous step, the computer device further refines its boundaries. Since the surface is composed of discrete triangular meshes, boundary identification is achieved by examining the edges within the mesh. If an edge is possessed by only one triangle, then that edge is part of the surface patch boundary. By traversing all triangles in the surface patch, identifying all such boundary edges, and then connecting them in an end-to-end order, a closed boundary contour is formed. This contour precisely defines the coverage area of ​​the power response region corresponding to the energy efficiency state node on the surface; that is, all mesh vertices and triangles within this contour belong to this node.

[0090] Step 155 involves calculating the power consumption contribution and analyzing the time interval for each surface segmentation sub-region to generate a set of power consumption distribution characteristic parameters. This step is achieved through the following sub-steps:

[0091] Step 1551: Perform regional volume integration on each surface segmentation sub-region, calculate the spatial volume enclosed between each surface segmentation sub-region and the base plane of the power consumption response spatiotemporal mapping surface structure, and obtain the regional power consumption contribution integral value corresponding to each energy efficiency state transition node.

[0092] For each segmented curved sub-region, the computer device calculates the volume of the space enclosed by it and the base plane. The base plane can be defined as the plane with a Y-coordinate of zero, i.e., the plane with zero power consumption. In the discretized triangular mesh representation, this volume can be approximated as the sum of the volumes of all triangular prisms projected vertically onto the base plane with the triangles on the curved surface as vertices. Specifically, for each triangle in the curved patch, its three vertices are projected vertically downwards onto the base plane, resulting in three projection points. These three projection points, together with the original three vertices, form a triangular prism. The volume of this triangular prism can be obtained by multiplying the area of ​​the triangle by the average Y-coordinate of the three vertices. The total volume integral value obtained by summing the calculated volumes of the triangular prisms for all triangles quantifies the total power consumption contributed by this energy efficiency state throughout the process.

[0093] Step 1552: Generate the power consumption contribution factor of each energy efficiency state transition node based on the regional power consumption contribution integral value corresponding to each energy efficiency state transition node. The power consumption contribution factor is used to represent the power consumption contribution of the energy efficiency state transition node on the energy efficiency state transition path.

[0094] The computer device directly uses the volume integral value calculated in step 1551 as the power consumption contribution factor of the node. The power consumption contribution factor is a dimensionless scalar, or it can be understood as a value with energy dimensions. It comprehensively reflects the duration of the state and the magnitude of power consumption, and is a key indicator for measuring the proportion of the state in the overall energy consumption.

[0095] Step 1553: Calculate the difference between the time coordinate components of the transition node coordinates of two adjacent energy efficiency state transition nodes, subtract the time coordinate component of the previous energy efficiency state transition node from the time coordinate component of the latter energy efficiency state transition node, and obtain the time interval value between adjacent energy efficiency state transition nodes, and use the time interval value as the time interval factor between transition nodes.

[0096] The computer device extracts the X-coordinates, i.e., the time coordinates, of two adjacent nodes. For nodes arranged in path order, such as nodes A and B, the X-component of the transition node coordinates of node A is denoted as X_A, and the X-component of node B is denoted as X_B. The calculation time interval is X_B minus X_A. This difference represents the time taken to switch from the energy efficiency state represented by node A to the energy efficiency state represented by node B, including the time spent on the transient transition process between the two states. This value is recorded as the time interval factor between transition nodes.

[0097] Step 1554: Associate and store the power consumption contribution factor corresponding to each energy efficiency state transition node and the time interval factor between transition nodes between each two adjacent energy efficiency state transition nodes to form a power consumption distribution characteristic parameter entry with energy efficiency state transition nodes as the unit. Arrange all power consumption distribution characteristic parameter entries corresponding to all energy efficiency state transition nodes according to the order of the energy efficiency state transition nodes on the energy efficiency state transition path to generate a power consumption distribution characteristic parameter set.

[0098] The computer device creates a data entry for each node. For example, for the first node on the path, this entry includes at least the node identifier and the node's power consumption contribution factor. Simultaneously, for each pair of adjacent nodes, an association entry is created, which includes at least the predecessor node identifier, the successor node identifier, and the time interval factor between the two nodes during the transition. For ease of organization, the time interval factor can also be included as part of the successor node entry. For example, for node B, its entry could include the node identifier, its own power consumption contribution factor, and the time interval factor for transitioning from predecessor node A to this node. Finally, these node entries are arranged into an ordered list according to the nodes' order on the path, forming a power consumption distribution characteristic parameter set. This set details the power consumption and time characteristics of each stage in the energy efficiency state transition path.

[0099] Step 160: Generate a power consumption test analysis conclusion identifier that includes an energy efficiency level identifier based on the power consumption distribution characteristic parameter set.

[0100] The computer equipment ultimately rates the overall energy efficiency of the LPDDR chip under test based on the extracted set of power consumption distribution characteristic parameters, and generates the final analysis conclusion. This step is achieved through the following methods:

[0101] Step 161: Obtain multiple power distribution characteristic parameter entries contained in the power distribution characteristic parameter set. Each power distribution characteristic parameter entry contains a power contribution factor corresponding to an energy efficiency state transition node and a time interval factor between the energy efficiency state transition node and the next energy efficiency state transition node.

[0102] The computer device extracts information about all nodes from the set of power consumption distribution characteristic parameters. For each node on the path, the power consumption contribution factor of that node is parsed from its corresponding entry. Simultaneously, the time interval factor required to transition from that node to the next node is parsed from that entry or associated entry. For the last node on the path, since it has no subsequent nodes, its inter-node time interval factor can be set to a null value or zero.

[0103] Step 162: Vectorize and encapsulate the power consumption contribution factor and the time interval factor between transfer nodes in each power consumption distribution characteristic parameter entry to generate a node feature vector corresponding to each energy efficiency state transfer node. The node feature vector has the power consumption contribution factor as the first component and the time interval factor between transfer nodes as the second component.

[0104] For each node on the path, the computer device uses its own power consumption contribution factor as the first element of a vector, and its transition time interval factor to the next node as the second element, combining them into a two-dimensional feature vector. For the last node, its time interval factor is set to zero. Thus, for a path containing N nodes, N two-dimensional feature vectors can be obtained.

[0105] Step 163: Serialize and arrange the node feature vectors corresponding to all energy efficiency state transition nodes according to the order of the energy efficiency state transition nodes on the energy efficiency state transition path to obtain a node feature vector sequence composed of multiple node feature vectors in sequence.

[0106] The computer device arranges these two-dimensional vectors strictly according to the order in which the nodes appear on the path, forming an ordered sequence called the node feature vector sequence. The length of the sequence is equal to the total number of nodes on the path. The node feature vector sequence organizes the local information of each node on the path into a whole.

[0107] Step 164: Perform sequence statistical feature extraction processing on the node feature vector sequence, calculate the arithmetic mean of the first component of all node feature vectors in the node feature vector sequence as the path average power consumption contribution feature value, and calculate the arithmetic mean of the second component of all node feature vectors in the node feature vector sequence as the path average time interval feature value.

[0108] The computer equipment performs global statistics on the node feature vector sequence. First, it extracts the first component of each vector in the sequence, which is the power consumption contribution factor of all nodes. These values ​​are summed and then divided by the total number of nodes. The resulting arithmetic mean is the path average power consumption contribution feature value. This feature value reflects the average power consumption level contributed by each energy efficiency state throughout the entire test. Second, it extracts the second component of each vector in the sequence, which is the time interval factor between all transition nodes. These values ​​are also summed and then divided by the number of effective intervals, i.e., the total number of nodes minus one. The resulting arithmetic mean is the path average time interval feature value, which reflects the average speed of state transitions.

[0109] Step 165: Perform sequence change feature extraction processing on the node feature vector sequence, calculate the absolute value of the difference between the first components of the feature vectors of adjacent nodes as the first component change sequence, calculate the absolute value of the difference between the second components of the feature vectors of adjacent nodes as the second component change sequence, sum the first component change sequence to obtain the path power consumption fluctuation cumulative feature value, and sum the second component change sequence to obtain the path time fluctuation cumulative feature value.

[0110] The computer equipment further analyzes the volatility within the sequence. For the first component, the power consumption contribution factor, starting from the first vector of the sequence, the absolute value of the first component of the second vector minus the absolute value of the first component of the first vector is calculated to obtain the first change. Then, the absolute value of the first component of the third vector minus the absolute value of the first component of the second vector is calculated to obtain the second change. This process continues until the (N-1)th change is calculated. All these changes constitute the first component change sequence. The sum of all changes in this sequence is the cumulative characteristic value of path power consumption volatility. This value reflects the sum of the differences in energy consumption levels between states along the entire energy efficiency state transition path; the greater the difference, the greater the cumulative volatility value. For the second component, the time interval factor, the same method is used: the absolute value of the difference between the second components of adjacent vectors is calculated to obtain the second component change sequence, and this sequence is summed to obtain the cumulative characteristic value of path time volatility. This value reflects the stability of the state switching rhythm.

[0111] Step 166: Perform feature splicing and fusion processing on the path average power consumption contribution feature value, the path average time interval feature value, the path power consumption fluctuation cumulative feature value, and the path time fluctuation cumulative feature value to generate the overall path feature vector corresponding to the energy efficiency state transition path.

[0112] The computer device concatenates the four scalar feature values ​​obtained from the above calculations in a fixed order to form a four-dimensional feature vector. The first component of the vector is the path average power consumption contribution feature value, the second component is the path average time interval feature value, the third component is the path power consumption fluctuation cumulative feature value, and the fourth component is the path time fluctuation cumulative feature value. This four-dimensional vector is a highly abstract comprehensive fingerprint that can fully characterize the energy efficiency behavior of the LPDDR chip under test throughout the entire testing process.

[0113] Step 167: Based on the overall feature vector of the path, perform energy efficiency level matching and generate and store conclusion identifiers. This step is implemented through the following sub-steps:

[0114] Step 1671: Call the pre-built standard energy efficiency level library. The standard energy efficiency level library contains multiple standard energy efficiency level path entries. Each standard energy efficiency level path entry corresponds to a standard energy efficiency level identifier and contains a standard path feature vector corresponding to that standard energy efficiency level path.

[0115] The computer accesses a pre-generated database, namely the energy efficiency level standard path library, which stores standards summarized from the testing and analysis of a large number of standard samples or historical batches of LPDDR particles. When constructing this library, a full-process analysis as described in steps 110 to 160 is first performed on a batch of particles with known energy efficiency levels to obtain the overall path feature vector for each particle. Then, the feature vectors of all particles of the same energy efficiency level are clustered or averaged to obtain a standard path feature vector that represents that level. Each standard energy efficiency level path entry contains this standard feature vector and its corresponding unique energy efficiency level identifier, such as Level 1, Level 2, Level 3, etc.

[0116] Step 1672: Calculate the vector similarity between the overall feature vector of the path and the standard path feature vector of each standard energy efficiency level path entry in the standard energy efficiency level path library, and obtain the path matching similarity corresponding to each standard energy efficiency level path entry.

[0117] The computer equipment calculates the similarity between the overall path feature vector of the particle under test and each standard feature vector in the database. The similarity calculation can be based on Euclidean distance. Specifically, the magnitude of the difference vector between two four-dimensional vectors is calculated; that is, the square of the difference between corresponding components is calculated first, then summed, and finally the square root is taken to obtain the Euclidean distance. The smaller this distance, the more similar the two vectors are. To obtain an intuitive similarity, the distance can be mapped to a range of 0 to 1 using a negative correlation function. For example, the similarity is equal to the reciprocal of the Euclidean distance plus 1 and then the reciprocal, or equal to 1 divided by 1 plus the Euclidean distance. Calculations are performed on all standard entries to obtain a path matching similarity set, where each element in the set corresponds to a matching similarity of a standard energy efficiency level.

[0118] Step 1673: Based on the path matching similarity, select the standard energy efficiency level path entry with the highest path matching similarity from the energy efficiency level standard path library as the target matching level path entry, and extract the standard energy efficiency level identifier corresponding to the target matching level path entry.

[0119] The maximum value in the path matching similarity set is identified. The standard energy efficiency level path entry corresponding to this maximum value is the standard level that most closely matches the energy efficiency behavior of the particle being tested. The computer device determines this entry as the target matching level path entry and extracts its associated stored standard energy efficiency level identifier.

[0120] Step 1674: The extracted standard energy efficiency level identifier is associated and encapsulated with the particle identifier information of the LPDDR chip under test to generate a power consumption test analysis conclusion identifier containing the particle identifier information and the standard energy efficiency level identifier.

[0121] The computer equipment acquires the unique identification information of the LPDDR chip under test, such as its serial number or batch number. Then, the chip identifier and the energy efficiency rating identifier extracted in step 1673 are encapsulated together according to a predetermined data format to form a complete conclusion identifier. For example, the two can be concatenated into a single string using a set delimiter.

[0122] Step 1675: Write the power consumption test analysis conclusion identifier into the designated storage area of ​​the target test data storage location, and update the index record of the target test data storage location.

[0123] Finally, the computer device writes the conclusion identifier into a designated field in the test results database and updates the database indexes for querying, statistical analysis, and product grading.

[0124] In an optional embodiment, the method further includes:

[0125] Step 210: Perform topological structure analysis on the power consumption distribution feature parameter set, extract energy efficiency state transition relationship features, determine the power consumption contribution factor of each energy efficiency state transition node and the time interval factor between adjacent nodes based on the energy efficiency state transition relationship features, and construct an energy efficiency state transition directed graph with nodes as graph nodes and transition relationships as directed edges, based on the order of each node in the energy efficiency state transition path descriptor, using the power consumption contribution factor as the node weight and the time interval factor as the edge weight.

[0126] In addition to generating the final conclusion identifier, this application embodiment can also perform deeper topological analysis on the power consumption distribution characteristic parameter set. The computer device first parses the set, identifying the nodes and edges. A node is a record of each energy efficiency state transition node in the set, and each node has a unique identifier. An edge represents the transition relationship between nodes, implicitly defined by the time interval factor recorded in the set, with the direction from the predecessor node to the successor node. Then, each node is assigned a weight, i.e., the node's power consumption contribution factor. Each directed edge is assigned a weight, i.e., the time interval factor between the transition nodes connecting the two nodes. Based on this information, the device constructs a directed graph structure. This directed graph consists of four parts: a node set, a directed edge set, a node weight mapping, and an edge weight mapping. The node set contains all nodes on the path. The directed edge set contains all ordered pairs, representing transitions from one node to another. The node weight mapping maps each node to its power consumption contribution factor. The edge weight mapping maps each directed edge to its time interval factor.

[0127] Step 220: Perform node centrality measurement on the directed graph, calculate the betweenness centrality value and degree centrality value of each node, calculate the clustering coefficient of the directed graph, and generate the clustering coefficient of each node based on the ratio of the actual number of edges in the set of adjacent nodes to the maximum number of candidate edges.

[0128] Based on the directed graph, the computer device performs a series of graph theory analyses. First, it calculates the degree centrality of each node. Degree centrality is divided into in-degree and out-degree. In-degree refers to the number of edges pointing to that node, and out-degree refers to the number of edges originating from that node and pointing to other nodes. The total degree of a node is the sum of its in-degree and out-degree. Degree centrality reflects the extent to which a state is connected to other states during energy efficiency transitions. Next, it calculates betweenness centrality. Calculating betweenness centrality requires finding the shortest paths between all pairs of nodes in the graph. For each node, it counts how many of the shortest paths between all pairs of nodes pass through that node. The proportion of this number of paths to the total number of shortest paths is the betweenness centrality of that node. Nodes with high betweenness centrality act as bridges in state transitions. Simultaneously, it calculates the clustering coefficient of each node. The clustering coefficient measures the tightness of the connections between a node's neighbors. For a node, it first finds all its neighbors, i.e., all nodes directly connected to that node. Then, it counts the actual number of edges between these neighbors. Finally, the ratio obtained by dividing the actual number of edges by the maximum possible number of edges between these neighboring nodes is the clustering coefficient of that node.

[0129] Step 230: The betweenness centrality value, degree centrality value and clustering coefficient of each node are fused to generate the node topology feature vector corresponding to each node. All node topology feature vectors are arranged in the order of nodes to obtain the node topology feature vector sequence.

[0130] For each node, its calculated betweenness centrality, degree centrality, and clustering coefficient are combined in a fixed order to form a three-dimensional topological feature vector. The first component of the vector is the betweenness centrality, the second is the degree centrality, and the third is the clustering coefficient. Then, according to the order of the nodes on the original energy efficiency state transition path, these three-dimensional vectors are arranged into a sequence, namely the node topological feature vector sequence.

[0131] Step 240: Perform graph topology similarity matching between the sequence and the standard node topology feature vector sequence of each standard graph in the pre-stored standard energy efficiency state transition graph library. Calculate the graph edit distance between the sequences and generate graph topology matching similarity to obtain a graph topology matching similarity set. Select the standard graph corresponding to the graph topology matching similarity with the largest value, extract the energy efficiency state transition mode identifier it carries, and attach the mode identifier as the energy efficiency state transition mode analysis conclusion to the power consumption test analysis conclusion identifier.

[0132] The computer equipment compares the sequence of topological feature vectors of the nodes to be tested with each standard graph in a standard graph library. The standard graph library stores graph structures of various known and representative energy efficiency state transition modes, along with the standard topological feature vector sequence of each node in these graph structures. During the comparison, a graph edit distance algorithm is used. Graph edit distance refers to the minimum number of operations or cost required to transform one graph into another through the insertion, deletion, and replacement of nodes and edges. In the calculation, not only the graph structure but also the topological feature vectors attached to the nodes are considered. If the vectors of two nodes are similar, the replacement cost is small; otherwise, the replacement cost is large. Using a dynamic programming algorithm, the minimum edit cost between the graph corresponding to the test sequence and the standard graph can be calculated; the reciprocal or negative correlation value of this cost is the graph topological matching similarity. After finding the standard graph that is most similar to the test sequence, i.e., the one with the highest graph topological matching similarity, the energy efficiency state transition mode identifier carried by this graph is extracted, such as "continuous transition mode," "jump transition mode," or "cyclic transition mode." Finally, the pattern identifier is appended to the power consumption test analysis conclusion identifier generated in step 167 to form a dual conclusion that includes both energy efficiency level and behavior pattern.

[0133] In an optional embodiment, the method further includes:

[0134] Step 310: Evaluate the power consumption response timing stability of the power consumption distribution characteristic parameter set, generate a power consumption fluctuation stability index, obtain the power consumption contribution factor of each energy efficiency state transition node through the power consumption fluctuation stability index, and arrange them into a power consumption contribution factor sequence according to the node order.

[0135] In addition to topology analysis, time-series stability assessment can also be performed. The computer device extracts the power contribution factors of all nodes from the set of power distribution characteristic parameters, and arranges them into a sequence strictly according to the order of the nodes on the path. This sequence is the power contribution factor sequence, which is a one-dimensional time series that reflects the fluctuation of power contribution as state transitions.

[0136] Step 320: Perform differential processing on the sequence, calculate the absolute value of the difference between adjacent power consumption contribution factors to obtain the power consumption contribution fluctuation amplitude sequence, extract the arithmetic mean of the absolute values ​​of all differences in the fluctuation amplitude sequence as the average fluctuation amplitude feature value, and extract its maximum value as the maximum fluctuation amplitude feature value.

[0137] The power consumption contribution factor sequence is processed using first-order differencing. Specifically, starting from the second element of the sequence, the value of the current element is subtracted from the value of the previous element, and the absolute value is taken to obtain the first fluctuation amplitude. This process is repeated until the last fluctuation amplitude is calculated. All these fluctuation amplitudes constitute a new sequence, namely the power consumption contribution fluctuation amplitude sequence. Then, the arithmetic mean of all elements in this fluctuation amplitude sequence is calculated to obtain the average fluctuation amplitude characteristic value, which reflects the average drastic change in power consumption contribution between states. At the same time, the maximum value in the fluctuation amplitude sequence is identified to obtain the maximum fluctuation amplitude characteristic value, which reflects the most drastic change in power consumption contribution during the entire process.

[0138] Step 330: Obtain the time interval factors between adjacent nodes and arrange them into a time interval factor sequence according to the corresponding relationship. Perform a consistency measurement on the time interval factor sequence, calculate the standard deviation of all time interval factors as the time interval dispersion feature value, and calculate its range as the time interval fluctuation range feature value.

[0139] Similarly, time interval factors between all adjacent nodes are extracted from the power consumption distribution characteristic parameter set and arranged into a sequence according to the order of node transitions, i.e., the time interval factor sequence. This sequence reflects the temporal rhythm of state switching. A consistency measure is then applied to this sequence. First, the standard deviation of the sequence is calculated: first, the average of all time interval factors is calculated, then the square of the difference between each factor and the average is calculated, summed, averaged again, and finally the square root is taken. This standard deviation is the time interval dispersion characteristic value, which reflects the stability of the state switching interval; the smaller the standard deviation, the more regular the switching. Second, the range of the sequence is calculated: the maximum value in the sequence is subtracted from the minimum value to obtain the time interval fluctuation range characteristic value, which reflects the overall variation range of the switching interval.

[0140] Step 340: Concatenate the average fluctuation amplitude feature value, the maximum fluctuation amplitude feature value, the time interval dispersion feature value, and the time interval fluctuation range feature value to generate the initial power consumption fluctuation stability feature vector.

[0141] The computer device concatenates the four feature values ​​calculated in steps 320 and 330 into a four-dimensional vector in a fixed order. The first component of the vector is the average fluctuation amplitude feature value, the second component is the maximum fluctuation amplitude feature value, the third component is the time interval dispersion feature value, and the fourth component is the time interval fluctuation range feature value. This vector is the initial power consumption fluctuation stability feature vector.

[0142] Step 350: Call the pre-built power fluctuation stability level determination model to perform level mapping on the feature vector. The model includes multiple stability level determination threshold intervals and power fluctuation stability level identifiers corresponding to each interval.

[0143] The device invokes a pre-trained decision model. The model's construction process is as follows: First, a large amount of historical test data is collected, and an initial power consumption fluctuation stability feature vector is calculated for each sample. Then, experts label these samples with stability levels based on their actual performance, such as "high stability," "medium stability," and "low fluctuation." Next, classification algorithms, such as support vector machines or decision trees, are used to learn decision boundaries in the feature space that can distinguish between different levels. These boundaries are represented in the model as multiple non-overlapping threshold intervals, each interval corresponding to a defined stability level label.

[0144] Step 360: Match the feature vector with each threshold interval to determine its target interval and extract the corresponding target power consumption fluctuation stability level identifier. Associate the target level identifier as the power consumption fluctuation stability assessment result with the power consumption test analysis conclusion identifier.

[0145] The computer device inputs the initial power consumption fluctuation stability feature vector of the current test chip into the judgment model invoked in step 350. The model determines which threshold range the feature vector falls into based on its component values. For example, if the feature vector falls into the range representing "high stability," the model outputs the corresponding level identifier "high stability." Finally, the computer device also associates this target power consumption fluctuation stability level identifier with the final power consumption test analysis conclusion identifier, forming a more comprehensive conclusion identifier together with the previous energy efficiency level identifier and transfer mode identifier. This comprehensive identifier fully summarizes the energy efficiency level, behavior pattern, and stability characteristics exhibited by the tested LPDDR chip in this test.

[0146] As can be seen, this application embodiment achieves a systematic and in-depth analysis and quantitative rating of the power consumption characteristics of LPDDR chips by upscaling the original power consumption data to a three-dimensional surface and using an energy efficiency model for feature extraction, template matching, path analysis and multi-dimensional evaluation.

[0147] In detail, this embodiment of the application maps the collected raw power consumption response data sequence to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and the rate of change of power consumption value as the depth axis, constructing a spatiotemporal mapping surface structure for power consumption response. This achieves a fundamental transformation of the power consumption behavior of LPDDR particles from a one-dimensional time series to a three-dimensional geometric form, allowing the originally implicit transient impact characteristics and steady-state response characteristics to be explicitly separated and quantified in geometric space. Based on this, a pre-constructed LPDDR particle energy efficiency model is called to perform surface feature analysis processing on the spatiotemporal mapping surface structure for power consumption response. This enables the accurate extraction of transient power consumption impact response feature sequences and steady-state power consumption response feature sequences from the surface curvature distribution field, avoiding the identification ambiguity problem caused by feature aliasing in traditional time-domain or frequency-domain analysis. Furthermore, by performing template matching between the extracted transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence in the standard energy efficiency state response template library of the LPDDR particle energy efficiency model, an energy efficiency state transition path descriptor is generated for the LPDDR particle under test to switch from the current energy efficiency state to the target energy efficiency state. This descriptor completely describes the complete path of energy efficiency state change within the particle in a structured form, providing a clear navigation basis for refined analysis.

[0148] Subsequently, the power consumption response spatiotemporal mapping surface structure is segmented according to the energy efficiency state transition path descriptor to obtain a set of power consumption distribution characteristic parameters for each node position on the energy efficiency state transition path of the LPDDR chip under test. This set quantifies and locally represents the power consumption energy contributed by each energy efficiency state and the time cost of state switching, making it possible to trace and allocate power consumption contributions. Finally, a power consumption test analysis conclusion identifier containing an energy efficiency level identifier is generated based on the set of power consumption distribution characteristic parameters. This achieves fully automated mapping from raw power consumption data to final energy efficiency level determination. The entire method is interconnected and progressive, constructing a complete systematic solution for power consumption feature space transformation, geometric feature extraction, state path identification, regional energy segmentation, and energy efficiency level assessment. This improves the accuracy, comprehensiveness, and automation level of LPDDR chip power consumption test analysis, thereby accurately realizing the energy efficiency classification and performance evaluation of memory chips.

[0149] Those skilled in the art will understand that mapping one-dimensional time-series power consumption data to three-dimensional space to construct a surface is essentially about giving the third dimension physical meaning. In this application's embodiments, the third dimension is explicitly defined as the rate of change of power consumption (mW / s), providing a clear physical basis for the construction of the three-dimensional space. Based on this, constructing a continuous surface from discrete points is a common operation in computer graphics and reverse engineering. Those skilled in the art can use point cloud data and existing technologies such as Poisson surface reconstruction or moving cube algorithms, combined with adaptive adjustment of smoothing parameters based on local curvature, to generate a smooth manifold surface that reflects the characteristics of the original data. As for feature extraction methods such as surface curvature calculation and extreme point clustering, these are fundamental tools in differential geometry and data mining. Technicians are capable of selecting appropriate algorithms based on the quality of the surface mesh (e.g., using local fitting methods to estimate principal curvature and employing the DBSCAN algorithm for spatial clustering), thereby transforming the functional descriptions in this application's embodiments into concrete technical implementations.

[0150] This application describes a progressive relationship between feature extraction and template matching, which is typical logic in pattern recognition and does not contain inherent contradictions. Potential misunderstandings may arise from the uniformity of feature vector dimensions and the applicability of the matching algorithm. For example, transient feature points are three-dimensional vectors, while steady-state feature intervals are quadruples; both require weighted processing during subsequent fusion. Those skilled in the art understand that when constructing comprehensive matching similarity, features with different dimensions and physical meanings cannot be directly compared numerically. Instead, they are calculated separately with the standard template (e.g., DTW distance, interval overlap) before being weighted and fused without dimensions. This approach is standard practice in information fusion and effectively resolves the logical dilemmas caused by feature heterogeneity. Furthermore, the logic of backtracking from the energy efficiency state transition path descriptor to the surface for region segmentation relies on the alignment relationship established during template matching. Technicians can use dynamic programming backtracking or direct indexing of corresponding time tags to map abstract path nodes back to specific spatiotemporal coordinates. This process is logically clear and technically feasible.

[0151] This application employs units with clearly defined physical meanings at different processing stages. In the three-dimensional spatial construction, the X-axis represents seconds, the Y-axis represents milliwatts, and the Z-axis represents milliwatts per second, achieving independence of the physical meaning of each dimension. During feature extraction and matching, although the feature vectors are composed of components with different units, those skilled in the art, when processing multi-source data fusion, will naturally employ standardization or normalization preprocessing to map data with different units to the same scale range, thereby eliminating the influence of units and ensuring fairness in distance calculation or similarity measurement. For example, before calculating the Euclidean distance between transient feature vectors, the time, power consumption, and rate of change components are normalized respectively; when constructing the overall path feature vector, the statistical feature values ​​are also standardized. The above-mentioned normalization operation is a common technique in the fields of data analysis and pattern recognition; therefore, although the original data units differ, they are unified through standardization within the algorithm.

[0152] Please see Figure 2 The figure is a schematic diagram of the basic structure of a computer device 200 provided in an embodiment of this application. The computer device 200 includes: a processor 201; a storage device 202 on which a computer program 2020 is stored; and a network interface 203 for providing network communication functions. When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the LPDDR chip power consumption test and analysis methods based on the energy efficiency model.

[0153] Please see Figure 3 This application provides a functional block diagram of an LPDDR chip power consumption test and analysis device. The LPDDR chip power consumption test and analysis device includes:

[0154] The response data acquisition module is used to acquire the raw power consumption response data sequence generated when the LPDDR chip under test runs the test stimulus program under preset test conditions. The raw power consumption response data sequence includes multiple power consumption sampling point values ​​with timestamps acquired continuously.

[0155] The spatiotemporal mapping processing module is used to map the original power consumption response data sequence to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and power consumption value change rate as the depth axis, to generate the power consumption response spatiotemporal mapping surface structure of the LPDDR chip under test under the test conditions.

[0156] The surface feature parsing module is used to call the pre-built LPDDR particle energy efficiency model to perform surface feature parsing processing on the power consumption response spatiotemporal mapping surface structure, and extract transient power consumption impact response feature sequences and steady-state power consumption response feature sequences from the power consumption response spatiotemporal mapping surface structure;

[0157] The template matching processing module is used to perform template matching processing in the standard energy efficiency state response template library of the LPDDR particle energy efficiency model according to the transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence, and generate an energy efficiency state transition path descriptor for the LPDDR particle under test to switch from the current energy efficiency state to the target energy efficiency state.

[0158] The power consumption test and analysis module is used to perform surface region segmentation processing on the power consumption response spatiotemporal mapping surface structure according to the energy efficiency state transition path descriptor, to obtain the power consumption distribution characteristic parameter set of the LPDDR chip under test at each node position on the energy efficiency state transition path, and to generate a power consumption test and analysis conclusion identifier containing an energy efficiency level identifier based on the power consumption distribution characteristic parameter set.

[0159] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0160] Furthermore, it should be noted that this application also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of a computer device reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the computer device to perform the aforementioned... Figure 1 The methods described in the corresponding embodiments are already known, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program product embodiments related to this application, please refer to the description of the method embodiments of this application.

[0161] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. A method for testing and analyzing the power consumption of LPDDR chips based on an energy efficiency model, characterized in that, The method includes: The raw power consumption response data sequence is collected when the LPDDR chip under test runs the test stimulus program under preset test conditions. The raw power consumption response data sequence includes multiple power consumption sampling point values ​​with timestamps collected continuously. The original power consumption response data sequence is mapped to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and the rate of change of power consumption value as the depth axis, to generate the spatiotemporal mapping surface structure of the power consumption response of the LPDDR chip under test under the test conditions. The pre-built LPDDR particle energy efficiency model is invoked to perform surface feature analysis on the power consumption response spatiotemporal mapping surface structure, and the transient power consumption impulse response feature sequence and steady-state power consumption response feature sequence are extracted from the power consumption response spatiotemporal mapping surface structure. Based on the transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence, template matching processing is performed in the standard energy efficiency state response template library of the LPDDR particle energy efficiency model to generate an energy efficiency state transition path descriptor for the LPDDR particle under test to switch from the current energy efficiency state to the target energy efficiency state. The power consumption response spatiotemporal mapping surface structure is segmented according to the energy efficiency state transition path descriptor to obtain the power consumption distribution characteristic parameter set of the LPDDR chip under test at each node position on the energy efficiency state transition path, and a power consumption test analysis conclusion identifier containing the energy efficiency level identifier is generated based on the power consumption distribution characteristic parameter set. The step of segmenting the power response spatiotemporal mapping surface structure according to the energy efficiency state transition path descriptor to obtain a set of power distribution characteristic parameters of the LPDDR chip under test at each node position on the energy efficiency state transition path includes: parsing multiple energy efficiency state transition node identifiers contained in the energy efficiency state transition path descriptor, where each energy efficiency state transition node identifier corresponds to a key position point on the energy efficiency state transition path; and querying the three-dimensional spatial coordinate position corresponding to each energy efficiency state transition node identifier from the power response spatiotemporal mapping surface structure to obtain the transition node of each energy efficiency state transition node. Point coordinates; taking the coordinates of each energy efficiency state transition node as the center, delineate the surface region segmentation range centered on the coordinates of the transition node on the power consumption response spatiotemporal mapping surface structure, obtaining the surface segmentation sub-region corresponding to each energy efficiency state transition node; perform region boundary identification processing on the surface segmentation sub-region corresponding to each energy efficiency state transition node, extract the boundary contour line of each surface segmentation sub-region, and determine the coverage range of each surface segmentation sub-region on the power consumption response spatiotemporal mapping surface structure based on the boundary contour line; perform power consumption contribution calculation and time interval analysis on each surface segmentation sub-region to generate a set of power consumption distribution characteristic parameters; The step of calculating the power consumption contribution and analyzing the time interval for each surface segmentation sub-region to generate a set of power consumption distribution characteristic parameters includes: performing regional volume integration on each surface segmentation sub-region to calculate the spatial volume enclosed between each surface segmentation sub-region and the base plane of the power consumption response spatiotemporal mapping surface structure, obtaining the regional power consumption contribution integral value corresponding to each energy efficiency state transition node; generating a power consumption contribution factor for each energy efficiency state transition node based on the regional power consumption contribution integral value corresponding to each energy efficiency state transition node, wherein the power consumption contribution factor is used to represent the power consumption contribution of the energy efficiency state transition node on the energy efficiency state transition path; and calculating the time coordinate components of the transition node coordinates of two adjacent energy efficiency state transition nodes. A difference calculation is performed by subtracting the time coordinate component of the previous energy efficiency state transition node from the time coordinate component of the subsequent energy efficiency state transition node to obtain the time interval value between adjacent energy efficiency state transition nodes. This time interval value is used as the time interval factor between transition nodes. The power consumption contribution factor corresponding to each energy efficiency state transition node and the time interval factor between every two adjacent energy efficiency state transition nodes are associated and stored to form a power consumption distribution characteristic parameter entry with energy efficiency state transition nodes as the unit. All power consumption distribution characteristic parameter entries corresponding to energy efficiency state transition nodes are arranged according to the order of the energy efficiency state transition nodes on the energy efficiency state transition path to generate a power consumption distribution characteristic parameter set.

2. The method according to claim 1, characterized in that, The step of mapping the original power consumption response data sequence to a three-dimensional space with time as the horizontal axis, power consumption value as the vertical axis, and the rate of change of power consumption value as the depth axis, to generate the spatiotemporal mapping surface structure of the power consumption response of the LPDDR chip under test under the test conditions includes: The time axis coordinate allocation process is performed on the value of each power consumption sampling point in the original power consumption response data sequence, and the timestamp corresponding to each power consumption sampling point value is converted into the time coordinate component value in the three-dimensional space to obtain the time coordinate positioning point corresponding to each power consumption sampling point value. The power consumption sampling point value in the original power consumption response data sequence is processed by power consumption numerical axis coordinate allocation, and each power consumption sampling point value is used as a power consumption numerical coordinate component value in the three-dimensional space to obtain the power consumption numerical coordinate positioning point corresponding to each power consumption sampling point value. The change rate calculation is performed on the values ​​of adjacent power consumption sampling points in the original power consumption response data sequence. The difference between each power consumption sampling point value and the previous power consumption sampling point value is calculated, and the difference is divided by the interval of the timestamps corresponding to the adjacent power consumption sampling point values ​​to obtain the power consumption value change rate corresponding to each power consumption sampling point value. The power consumption value change rate is used as the change rate coordinate component value in the three-dimensional space to obtain the change rate coordinate positioning point corresponding to each power consumption sampling point value. The time coordinate positioning point, the power value coordinate positioning point and the rate of change coordinate positioning point corresponding to each power consumption sampling point value are processed to construct three-dimensional spatial points, thereby generating a spatial coordinate point representation of each power consumption sampling point value in the three-dimensional space. The spatial coordinate points corresponding to all power consumption sampling points in the original power consumption response data sequence are sequentially connected. Adjacent spatial coordinate points are connected by straight line segments according to the order of the time coordinate positioning points to obtain the power consumption response spatial trajectory line of the original power consumption response data sequence in the three-dimensional space. Based on the power response spatial trajectory line, surface expansion and smoothing are performed to construct a power response spatiotemporal mapping surface structure.

3. The method according to claim 1, characterized in that, The pre-built LPDDR particle energy efficiency model is invoked to perform surface feature analysis on the power response spatiotemporal mapping surface structure, extracting transient power surge response feature sequences and steady-state power response feature sequences from the power response spatiotemporal mapping surface structure, including: The power response spatiotemporal mapping surface structure is subjected to surface curvature calculation processing. The Gaussian curvature value and average curvature value of each surface grid vertex on the power response spatiotemporal mapping surface structure are calculated to obtain the surface curvature distribution field composed of the Gaussian curvature value and average curvature value of all surface grid vertices. The surface curvature distribution field is subjected to extreme point detection processing. The surface mesh vertices in the surface curvature distribution field with Gaussian curvature value or average curvature value greater than a preset curvature threshold are identified as candidate curvature extreme points. The three-dimensional spatial coordinate position of each candidate curvature extreme point on the power consumption response spatiotemporal mapping surface structure is recorded. Spatial clustering is performed on the candidate curvature extrema points to group candidate curvature extrema points with adjacent three-dimensional spatial coordinate positions into a curvature extrema point cluster. Each curvature extrema point cluster corresponds to a transient power consumption impact event region. Cluster center coordinates are calculated for each curvature extrema point cluster. The arithmetic mean of the three-dimensional spatial coordinate positions of all candidate curvature extrema points in each curvature extrema point cluster is used as the cluster center coordinates of the curvature extrema point cluster, and the cluster center coordinates are used as transient power consumption impact response feature points. The cluster center coordinates corresponding to all curvature extreme point clusters on the power consumption response spatiotemporal mapping surface structure are sorted according to the order of the time coordinate positioning points to generate a transient power consumption impulse response feature sequence composed of multiple transient power consumption impulse response feature points arranged in time order. Flat region detection and projection processing are performed on the spatiotemporal mapping surface structure of the power consumption response to extract the steady-state power consumption response feature sequence.

4. The method according to claim 3, characterized in that, The step of performing flat region detection and projection processing on the spatiotemporal mapping surface structure of the power consumption response to extract the steady-state power consumption response feature sequence includes: Flat region detection processing is performed on the power response spatiotemporal mapping surface structure, and continuous surface regions in the power response spatiotemporal mapping surface structure whose Gaussian curvature value and average curvature value are both less than a preset flat curvature threshold are identified as candidate flat regions. The candidate flat regions are subjected to regional connectivity analysis, and spatially connected candidate flat regions are merged into the same flat region unit. The range of three-dimensional spatial coordinates of each flat region unit on the power consumption response spatiotemporal mapping surface structure is recorded. Coordinate projection processing is performed on the three-dimensional spatial coordinate range corresponding to each flat region unit, and the three-dimensional spatial coordinate range of each flat region unit is projected onto a two-dimensional plane composed of the time axis and the power consumption value axis, so as to obtain the start time coordinate value and end time coordinate value of each flat region unit on the time axis, and the start power consumption coordinate value and end power consumption coordinate value on the power consumption value axis. Based on the start time coordinates, end time coordinates, start power consumption coordinates, and end power consumption coordinates of each flat region cell, a steady-state power consumption response characteristic interval is generated for each flat region cell. The steady-state power consumption response characteristic intervals of all flat region cells are then arranged in chronological order to generate a steady-state power consumption response characteristic sequence.

5. The method according to any one of claims 1-4, characterized in that, The step involves performing template matching processing on the standard energy efficiency state response template library of the LPDDR chip energy efficiency model based on the transient power consumption impulse response feature sequence and the steady-state power consumption response feature sequence to generate an energy efficiency state transition path descriptor for the LPDDR chip under test to switch from the current energy efficiency state to the target energy efficiency state, including: The three-dimensional spatial coordinates of each transient power consumption impulse response feature point in the transient power consumption impulse response feature sequence are converted into transient feature vectors with time coordinate components, power consumption numerical coordinate components, and rate of change coordinate components as elements, resulting in a transient feature vector sequence composed of multiple transient feature vectors; Each steady-state power response feature interval in the steady-state power response feature sequence is converted into a steady-state feature vector with the start time coordinate, end time coordinate, start power coordinate, and end power coordinate as elements, resulting in a steady-state feature vector sequence composed of multiple steady-state feature vectors; The standard energy efficiency state response template library stored in the LPDDR particle energy efficiency model is called. The standard energy efficiency state response template library contains multiple standard energy efficiency state response templates. Each standard energy efficiency state response template corresponds to a standard energy efficiency state type and contains a standard transient feature vector sequence and a standard steady-state feature vector sequence. The transient feature vector sequence is matched with the standard transient feature vector sequence of each standard energy efficiency state response template in the standard energy efficiency state response template library. The DTW distance between the transient feature vector sequence and each standard transient feature vector sequence is calculated, and a transient matching similarity is generated based on the DTW distance to obtain a transient matching similarity set composed of multiple transient matching similarities. The steady-state feature vector sequence is matched with the standard steady-state feature vector sequence of each standard energy efficiency state response template in the standard energy efficiency state response template library. The interval overlap ratio of each steady-state feature vector in the steady-state feature vector sequence and the corresponding standard steady-state feature vector in each standard steady-state feature vector sequence is calculated. The steady-state matching similarity is generated according to the interval overlap ratio, resulting in a steady-state matching similarity set composed of multiple steady-state matching similarities. The transient matching similarity set and the steady-state matching similarity set are weighted and fused. A first fusion weight coefficient is assigned to each transient matching similarity, and a second fusion weight coefficient is assigned to each steady-state matching similarity. The comprehensive matching similarity corresponding to each standard energy efficiency state response template is calculated. Based on the comprehensive matching similarity, the standard energy efficiency state response template with the highest comprehensive matching similarity is selected from the standard energy efficiency state response template library as the target matching template, and the standard energy efficiency state type identifier corresponding to the target matching template is extracted; Based on the standard energy efficiency state type identifier corresponding to the target matching template and the standard energy efficiency state transition relationship recorded in the target matching template, an energy efficiency state transition path descriptor is generated for the LPDDR chip under test to switch from the current energy efficiency state to the target energy efficiency state.

6. The method according to any one of claims 1-4, characterized in that, The generation of a power consumption test analysis conclusion identifier containing an energy efficiency level identifier based on the power consumption distribution characteristic parameter set includes: Obtain multiple power distribution feature parameter entries contained in the power distribution feature parameter set. Each power distribution feature parameter entry contains a power contribution factor corresponding to an energy efficiency state transition node and a time interval factor between the energy efficiency state transition node and the next energy efficiency state transition node. The power consumption contribution factor and the time interval factor between transfer nodes in each power consumption distribution characteristic parameter entry are vectorized and encapsulated to generate a node feature vector corresponding to each energy efficiency state transfer node. The node feature vector has the power consumption contribution factor as the first component and the time interval factor between transfer nodes as the second component. The node feature vectors corresponding to all energy efficiency state transition nodes are sequentially arranged according to the order of the energy efficiency state transition nodes on the energy efficiency state transition path, resulting in a node feature vector sequence composed of multiple node feature vectors in sequence. The node feature vector sequence is subjected to sequence statistical feature extraction processing. The arithmetic mean of the first component of all node feature vectors in the node feature vector sequence is calculated as the path average power consumption contribution feature value. The arithmetic mean of the second component of all node feature vectors in the node feature vector sequence is calculated as the path average time interval feature value. The node feature vector sequence is subjected to sequence change feature extraction processing. The absolute value of the difference between the first components of the feature vectors of adjacent nodes is calculated as the first component change sequence. The absolute value of the difference between the second components of the feature vectors of adjacent nodes is calculated as the second component change sequence. The first component change sequence is accumulated and summed to obtain the cumulative feature value of path power consumption fluctuation. The second component change sequence is accumulated and summed to obtain the cumulative feature value of path time fluctuation. The path average power consumption contribution feature value, the path average time interval feature value, the path power consumption fluctuation cumulative feature value, and the path time fluctuation cumulative feature value are subjected to feature splicing and fusion processing to generate the overall path feature vector corresponding to the energy efficiency state transition path. Based on the overall feature vector of the path, energy efficiency level matching, conclusion label generation, and storage are performed.

7. The method according to claim 6, characterized in that, The process of matching energy efficiency levels and generating and storing conclusion identifiers based on the overall feature vector of the path includes: Call the pre-built standard energy efficiency level library, which contains multiple standard energy efficiency level path entries. Each standard energy efficiency level path entry corresponds to a standard energy efficiency level identifier and contains a standard path feature vector corresponding to that standard energy efficiency level path. Calculate the vector similarity between the overall feature vector of the path and the standard path feature vector of each standard energy efficiency level path entry in the standard energy efficiency level path library to obtain the path matching similarity corresponding to each standard energy efficiency level path entry. Based on the path matching similarity, the standard energy efficiency level path entry with the highest path matching similarity is selected from the standard energy efficiency level path library as the target matching level path entry, and the standard energy efficiency level identifier corresponding to the target matching level path entry is extracted. The extracted standard energy efficiency level identifier is associated and encapsulated with the particle identifier information of the LPDDR chip under test to generate a power consumption test analysis conclusion identifier containing the particle identifier information and the standard energy efficiency level identifier. Write the power consumption test analysis conclusion identifier into the designated storage area of ​​the target test data storage location, and update the index record of the target test data storage location.

8. A computer device, characterized in that, include: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the LPDDR chip power consumption test and analysis method based on the energy efficiency model as described in any one of claims 1-7.

Images