Low on-resistance MOSFET system based on chip technology

By coupling the power MOS chip model and the package conductive network model in a low on-resistance MOS transistor system, a distributed on-resistance descriptor is generated, which solves the cross-scale correspondence problem between the device and package models, optimizes chip assembly and package interconnection, reduces the equivalent on-resistance after packaging, and improves calculation accuracy and consistency.

CN123117007APending Publication Date: 2026-11-17SHENZHEN XINTONG ELECTRONIC TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611305796.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies lack cross-scale correspondence between device models and packaging models of low on-resistance power metal-oxide-semiconductor field-effect transistors (MOS transistors), resulting in ineffective feedback of current migration and heat source migration, leading to local current accumulation, increased temperature rise, and increased on-resistance.

Method used

A low on-resistance MOSFET system based on chip technology is adopted. By coupling the power MOSFET chip model and the package conductive network model, a distributed conduction response descriptor is generated to record the current and heat source distribution. The interface sub-region boundary is fed back in the package branch to form a cross-scale conduction chain, identify and adjust mismatched edges, and optimize chip combination and package interconnection.

Benefits of technology

It effectively reduces the equivalent on-resistance after packaging, reduces local current density and heat source intensity, maintains current continuity, improves the accuracy and consistency of packaging calculations, and adapts to changes in process structure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117007A_ABST
    Figure CN123117007A_ABST
Patent Text Reader

Abstract

This invention relates to the field of computer-aided design and co-simulation of semiconductor power devices and chip packaging, specifically a low on-resistance MOSFET system based on chip technology. The system divides the MOSFET chip electrodes into interface sub-regions and generates distributed conduction response descriptors that record current and heat source migration based on potential and temperature boundary changes. According to the chip pose and contact coverage relationship, the interface nodes are mapped to the package conductive network branches. By alternately solving the device model and the package model, boundary, current, and loss data are returned, identifying conduction mismatches between internal high-current response paths and high-potential-drop branches in the package, and adjusting the chip pose and package contact boundaries. The system forms a closed-loop correlation between the internal conduction distribution of the device and the package interconnection state, reducing current congestion and heat source concentration, suppressing on-resistance growth caused by temperature rise, maintaining a low system equivalent on-resistance for the chip assembly, and ensuring consistent design results under different process structures.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of computer-aided design and chip-packaging co-simulation of semiconductor power devices, specifically to a low on-resistance MOS transistor system based on chip technology. Background Technology

[0002] Existing low on-resistance power metal-oxide-semiconductor field-effect transistors (MOSFETs) typically employ device-level computer-aided simulation (CADS) for structural and process optimization. Designers create planar, trench, or superjunction device models based on requirements for voltage withstand capability, current carrying capacity, junction temperature, and chip area. Structural information such as gate oxide layer, cell spacing, trench morphology, drift region doping, substrate, and metal contacts is written into the device mesh. By solving for potential distribution, carrier transport, and heat conduction, output characteristics, breakdown state, local temperature rise, and resistance contribution of each conductive region are obtained. This is then combined with ion implantation, diffusion, etching, and deposition processes to form the process structure, and the simulation model is calibrated using sample testing results. When the target device consists of multiple power chips, existing designs often optimize each chip individually before creating the package layout. Each chip is simplified to a concentrated resistor and heat source. Redistribution lines, copper clips, bonding structures, and external power terminals are connected according to the chip's location. Branch voltage drops, parallel currents, and junction temperatures are calculated using the package circuit and thermal network. Based on the calculation results, the number of chips, their placement, or the shape of the package conductors are then modified.

[0003] The above-mentioned processing method typically only transmits centralized results such as overall on-resistance, total loss, and junction temperature between the device model and the package model. The device model uses uniform terminal potential and idealized heat dissipation boundaries, while the package model does not retain the spatial correspondence between different regions of the chip electrode and the internal conduction path. When the chip changes position, orientation, or contact coverage within the package, the non-uniform potential and temperature boundaries formed by the package branch cannot be returned to the corresponding device mesh. The current migration and heat source migration generated inside the chip cannot be fed back to the specific package branch. This may cause internal regions with high current carrying capacity to connect to package paths that contribute significantly to potential drop, resulting in local current accumulation, increased temperature rise, and temperature-induced increase in on-resistance. Therefore, the technical problem to be solved is how to maintain the cross-scale correspondence between internal conduction regions, chip electrode interfaces, and package branches between device-level structural simulation and chip package simulation, and how to transform the low-resistance design result of a single chip into a low equivalent on-resistance chip system after packaging through bidirectional boundary transmission. Summary of the Invention

[0004] The purpose of this invention is to provide a low on-resistance MOSFET system based on chip technology, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A low on-resistance MOS transistor system based on chip technology includes a power MOS chip model and a packaged conductive network model coupled to the power MOS chip model. The power MOS chip model forms interface sub-regions according to the chip electrode regions, and generates distributed conduction response descriptors based on the current distribution and heat source distribution of each interface sub-region under the potential boundary and temperature boundary. The packaged conductive network model maps the distributed conduction response descriptors to the corresponding package branches according to the chip arrangement position and electrode connection relationship. It feeds back the interface sub-region boundary to the power MOS chip model and receives the returned interface sub-region current and local loss, so as to determine the chip combination and package interconnection scheme based on the feedback conduction response.

[0006] Preferably, the power MOS chip model discretizes the source metal projection boundary and the drain contact projection boundary into the interface sub-regions respectively, and merges the grid regions in the device grid that are located on the same continuous conduction path into the internal conduction region; The distributed conduction response descriptor adopts a directed graph. The interface node in the directed graph is connected to the corresponding internal region node. The response edge connecting the two is associated with the current migration record and heat source migration record before and after the boundary state change of the interface sub-region, and the index relationship between the interface node and the geometric position of the core electrode is retained.

[0007] Preferably, the packaged conductive network model defines the chip mounting area and external power terminals as network nodes, the continuous conductors connecting the network nodes as network branches, and establishes a pose mapping from chip local coordinates to package layout coordinates for each chip instance; The pose mapping projects the interface nodes in the distributed conduction response descriptor to the corresponding chip mounting area, establishes the interface node-network branch association according to the geometric coverage relationship between the projected interface boundary and the package contact boundary, and merges the association relationship sharing the same package contact boundary into the same current continuity constraint.

[0008] Preferably, the power MOS chip model and the packaged conductive network model are coupled according to an alternating solution sequence. The packaged conductive network model first forms the potential boundary and temperature boundary of each network node based on the existing interface sub-region current and writes them into the corresponding interface node. The power MOS chip model then updates the distributed conduction response descriptor based on the written boundary and returns the new interface sub-region current and local loss. Each round of coupling saves the mapping version between the interface node and the network branch, and the coupling ends when the branch current order and local loss position in adjacent mapping versions remain consistent.

[0009] Preferably, the internal conductive region is traced along the source-drain current direction by the device grid, and the region boundary is established at the position where the current direction bifurcates or the potential gradient contribution is transferred. Internal region nodes that start from the same interface node and are continuously connected to the drain contact boundary are sequentially connected to form an internal conduction chain. The response edge records the current migration record and heat source migration record of the boundary state change in the corresponding internal conduction chain. Each internal conduction chain is associated with the corresponding package branch through the interface node index.

[0010] Preferably, after the interface node is projected onto the encapsulation contact boundary, it is divided into covered sub-nodes and non-covered sub-nodes according to the overlapping grid of the projected boundary and the encapsulation contact boundary, and only the covered sub-nodes are connected to the corresponding network branches; The covered sub-nodes receive the package potential boundary according to the index relationship between the overlapping grid and the internal conductive chain. The temperature boundary formed by the chip mounting layer is written into the internal region node according to the same overlapping grid. The returned interface sub-region current flows into the network branch along the package contact boundary corresponding to the covered sub-node.

[0011] Preferably, after each round of coupling, each internal conduction chain and the network branch associated with its covered sub-node are connected in series to form a cross-scale conduction chain, and the internal current response order of the core and the contribution order of the unit current potential drop of the packaged branch are formed under the uniform terminal perturbation. The cross-scale conduction chains with common source network nodes and common drain network nodes are compared accordingly. The internal conduction chains with higher current response rankings are connected to network branches with higher potential drop contribution rankings and the association is marked as a conduction mismatch edge. A mismatch topology graph is formed based on the core instance or network branch shared by the conduction mismatch edge.

[0012] Preferably, the mismatched conduction edges in the mismatched topology diagram are used as the adjustment objects. While keeping the power MOS chip model itself unchanged, the pose mapping is first exchanged between chip instances with the same electrode definition, and the interface node-network branch association is migrated synchronously with the pose mapping. When the original mismatched conduction edge is still retained after the exchange, the package contact boundary is reconstructed along the adjacent network branches connecting the mismatched conduction edge in the mismatched topology graph. The reconstruction result is rewritten into the package conductive network model, and only the candidate package layouts of the mismatched conduction edge with the highest potential drop contribution position are retained.

[0013] Preferably, the power MOS chip model also receives a semiconductor process simulation structure corresponding to the candidate package layout, and projects the metal contact contour and doping boundary in the process simulation structure to the interface node and internal region node. When the node connection relationship before and after projection is consistent, the current migration record and heat source migration record of the corresponding response edge are updated with the device mesh after projection. When the node connection relationship changes, only the internal conduction chain crossed by the changed boundary is reconstructed, and coupling and the generation of the mismatched topology graph are re-executed on the updated distributed conduction response descriptor.

[0014] Preferably, for each mismatched topology diagram formed by simulation structures of different semiconductor processes, the candidate package layout is projected into a cross-process structure candidate set using the chip instance identifier and the interface node-network branch association as the alignment benchmark. The cross-process structure candidate set retains candidates whose conduction mismatch edges do not appear repeatedly in each mismatch topology diagram, and encapsulates the chip pose mapping, package contact boundary, and version relationship of the distributed conduction response descriptor corresponding to the retained candidates into low on-resistance MOSFET system design data.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The distributed conduction response descriptor maintains an indexed association between the chip electrode interface sub-region and the internal conduction region, and records the current migration and heat source migration corresponding to boundary state changes. The packaged conductive network maps the interface nodes to specific network branches based on the chip pose and contact coverage relationship, so that the local potential boundaries and temperature boundaries obtained from the package calculation can be written back to the corresponding device mesh. The interface current and local loss after device simulation updates can be returned to the same package branch. After forming a cross-scale conduction chain through alternating solutions, conduction mismatch edges can be identified based on the internal current response order and the unit current potential drop contribution order of the package branch. The mismatch association can be eliminated by exchanging the chip pose mapping or reconstructing the contact boundaries of adjacent packages. This connects the internal conduction chain with higher current response to the package branch with lower potential drop contribution, reduces the cancellation of the intrinsic low resistance result of the device by the package series resistance, reduces the local current density concentration and local heat source intensity, and suppresses the increase in conduction resistance caused by temperature rise. The equivalent conduction resistance after packaging is reduced from three continuous links: internal resistance of the device, additional resistance of the package, and temperature-related resistance.

[0016] 2. Interface nodes are divided into covered and non-covered sub-nodes based on the overlapping mesh of the projected boundary and the package contact boundary. Potential boundaries, temperature boundaries, and return currents are all propagated along the same geometric index, which avoids boundary mismatches caused by the concentrated equivalent of local contact areas and maintains the current continuity between the chip model and the package network. The metal contact contours and doped boundaries in the semiconductor process simulation structure are reprojected to the interface nodes and internal region nodes. When the node connection relationship remains unchanged, only the response edge record is updated. When the node connection relationship changes, only the internal conduction chain crossed by the changed boundary is reconstructed, so that the process structure changes can be propagated to the candidate package layout along the existing cross-scale mapping without the need to repeatedly construct unaffected conduction chains. The mismatch topology graphs formed by different process structures are aligned through chip instance identifiers and interface-branch associations, which can filter out package candidates that repeatedly have the same mismatch edges under multiple process structures, and save the version relationship between chip pose, package contact boundary, and response descriptor, thereby reducing the model update range and the amount of repeated solution, and maintaining the verifiability of candidate results and consistency across process structures. Attached Figure Description

[0017] Figure 1 A flowchart for cross-scale modeling and alternating coupling of device-package; Figure 2 Sorting graph of response edge coupling coefficients; Figure 3 The equivalent on-resistance convergence trend graph; Figure 4 This is a flowchart illustrating the process of identifying mismatches, adjusting package layouts, and screening across processes. Detailed Implementation

[0018] In one embodiment, a low on-resistance MOSFET system based on chip technology includes a power MOSFET chip model and a package conductive network model coupled to the power MOSFET chip model. The power MOSFET chip model forms interface sub-regions according to the chip electrode regions, and generates distributed conduction response descriptors based on the current distribution and heat source distribution of each interface sub-region under potential and temperature boundaries. The package conductive network model maps the distributed conduction response descriptors to the corresponding package branches according to the chip arrangement position and electrode connection relationship, feeds back the interface sub-region boundaries to the power MOSFET chip model, and receives the returned interface sub-region current and local losses, so as to determine the chip combination and package interconnection scheme based on the feedback conduction response.

[0019] In this embodiment, please refer to Figure 1The system is deployed on a computing platform used for device structure design and package layout design. The computing platform receives the device mesh file, material region identifiers, electrode boundary identifiers, process structure file, and package layout file of the power MOS chip, and unifies the geometric coordinates from different simulation environments into a two-layer coordinate system consisting of chip local coordinates and package global coordinates. The power MOS chip model saves the spatial position, material, adjacent meshes, and corresponding electrode region of each node in the device mesh. The package conductive network model saves the topological relationships between the chip mounting area, package contact boundary, continuous conductor, and external power terminals. The computing platform does not replace the entire chip with a single lumped resistor. Instead, the interface sub-regions after the electrode surface of each chip are divided are used as data exchange ports between the two types of models. The interface current and local loss generated on the device side are written to the package side through the port index, and the local potential and temperature generated on the package side are returned to the device side through the same port index. This forms a data closed loop that can maintain the correspondence between geometric position, conduction path and thermal boundary. The data objects output by the system include chip combination record, chip pose mapping, association between interface sub-region and package branch, coupling results of each round and candidate package layout. All data objects are associated with the same design task identifier so that the corresponding version can be called to continue the calculation when the device structure changes or the package contact boundary is rebuilt.

[0020] Specifically, the power MOS chip model is jointly formed by the device structure solution domain and the process structure solution domain. The device structure solution domain establishes a non-uniform grid according to the main current direction from the source to the drain. In the junction region, channel proximity region, drift region, substrate region, and metal contact proximity region, grid connections that can express potential gradients and heat source gradients are retained. The process structure solution domain generates material boundaries and doped boundaries based on implantation, diffusion, etching, deposition, and metallization processes, and maps the generated boundaries back to the device structure solution domain. During device solving, a conduction state is applied at the gate boundary, and an external state is applied at the source and drain boundaries. The potential conditions transmitted from the circuit are applied to the surface and back of the chip, and the temperature conditions transmitted from the packaging model are applied to the surface and back of the chip. The potential, electron current, hole current, Joule heat and heat flow direction at the grid nodes are obtained through the carrier continuity relationship, potential relationship and thermal conduction relationship. Then the results are integrated along the electrode projection range to form the current and local loss records of each interface sub-region. The interface sub-region records are not stored separately from the original device grid, but retain references to the set of grid nodes involved in the integration and their boundary surfaces, so that the package side can directly recalculate the affected area after returning to the new boundary without rebuilding the complete device structure.

[0021] Furthermore, the packaged conductive network model extracts the chip mounting area, chip top contact area, chip back contact area, redistribution conductor, copper clip connection, bonding connection, and external terminals from the package layout. During the extraction process, layout segments belonging to the same conductor are merged according to geometric continuity. Network nodes are established at conductor cross-section changes, material changes, connection bifurcation, and chip contact boundaries. Network branches are formed between adjacent network nodes. Each network branch stores layout geometric references, conductive material references, thermally conductive material references, and references to the boundary shared with adjacent branches. After the chip instance is placed in the chip mounting area, the system reads the chip electrode contour and executes coordinate... The transformation projects the interface sub-region onto the corresponding package contact boundary. The connection between the interface sub-region and the network branch is determined by the geometric overlap relationship after projection. The package electrical solver calculates the network node potential and branch potential drop based on the interface current. The package thermal solver calculates the core contact boundary temperature based on the local interface loss and the package conductor loss. The potential and temperature results are written back to the device model according to the projection index of the interface sub-region. After the device model is updated, the new current and new loss are written back to the package network. During the loop, the sign conventions of the current direction, heat flow direction and coordinate direction are kept fixed to avoid the data interpretation of the same interface in different simulation domains with opposite directions.

[0022] In a preferred embodiment, the power MOS chip model discretizes the source metal projection boundary and the drain contact projection boundary into interface sub-regions, and merges the grid regions located on the same continuous conduction path in the device grid into internal conduction regions. The distributed conduction response descriptor adopts a directed graph, in which the interface nodes in the directed graph are connected to the corresponding internal region nodes. The response edges connecting the two are associated with the current migration records and heat source migration records before and after the boundary state change of the interface sub-region, and the index relationship between the interface nodes and the geometric position of the chip electrode is preserved.

[0023] In this embodiment, the source metal projection boundary is discretized according to the intersection of the metal contact profile and the device grid surface. The system first closes the intersection to form several projection polygons, and then uses the device grid boundary surface covered by the projection polygon as the initial interface sub-region. When there is a current direction bifurcation, metal bridging boundary or package contact coverage breakpoint inside the same projection polygon, the initial interface sub-region continues to be split along the corresponding grid boundary, so that the internal current path connected by each interface sub-region remains continuous after entering the device. The drain contact projection boundary adopts the same geometric processing method, but its interface sub-region index and the source interface sub-region index use different prefixes to keep the source and drain port meanings fixed. The internal conductive region is merged starting from the device grid adjacent to the interface sub-region, and the adjacent grid with continuous current vector is accessed along the source and drain current direction. When there is a material boundary between adjacent grids but the current is continuous, they are retained in the same conductive region. When the current has a stable bifurcation or the local potential gradient changes, a new internal region node is established. The system saves the grid set, entry boundary, exit boundary and adjacent internal region node index for each internal region node, and constructs a directed connection pointing to the drain contact boundary starting from the interface node.

[0024] Specifically, the interface nodes of the distributed conduction response descriptor store the interface sub-region identifier, core electrode identifier, local coordinate boundary, and encapsulation projection reference. The internal region nodes store the mesh reference and connection order of the conduction region. The response edge stores the direction from one interface node or internal region node to the next internal region node, and records the current and heat source changes in the corresponding path before and after the boundary state change. During descriptor generation, a baseline solution, a potential boundary perturbation solution, and a temperature boundary perturbation solution are performed on the same core. The baseline solution provides the interface current and region heat source in the unperturbed state. The potential boundary perturbation solution... The perturbation only changes the potential input of the interface sub-region under investigation while keeping the other boundaries unchanged. The temperature boundary perturbation only changes the temperature input of the interface sub-region under investigation while keeping the potential conditions unchanged. Each solution result is projected onto the same internal region node according to the grid index. The system forms a response record based on the difference before and after the perturbation. If the perturbation of one interface sub-region causes the current redistribution of multiple internal regions, the interface node establishes response edges to the corresponding internal region nodes respectively. If multiple interface sub-regions affect the same internal region, the internal region node retains multiple incoming edges to express the common constraints of different positions of the core electrode on the internal conduction process.

[0025] Preferably, in order to enable quantitative comparison of the response edge between the device model and the package model, a dimensionless coupling coefficient can be used to represent the comprehensive influence of the interface sub-region on the internal conductive region. The coupling coefficient is calculated as follows.

[0026] ; in, This represents the coupling coefficient of the response edge between the i-th interface sub-region and the j-th internal conductive region. This represents the change in the potential boundary of the i-th interface sub-region. This represents the temperature boundary change within the same interface sub-region. This represents the change in current corresponding to the j-th internal conducting region. This represents the change in heat source power corresponding to the j-th internal conductive region. This represents the reference conductance used for normalizing the current-voltage response. This represents the reference thermal response coefficient used to normalize the temperature response of the heat source. and These represent the dimensionless weights of the electrical and thermal responses, respectively, and their sum is 1. In the example calculation, when the current and voltage responses of a certain response edge are 4 times the reference conductance and the heat source temperature response is 0.5 times the reference thermal response coefficient, and the electrical response weight is set to 0.75 and the thermal response weight is set to 0.25, the coupling coefficient is 3.125. If the coupling coefficient of another response edge obtained under the same normalization condition is 1.2, then the former response edge is arranged before the latter response edge in the descriptor. During the encapsulation mapping, it is first checked whether the interface sub-region corresponding to the former response edge is connected to the network branch with a large unit current potential drop. In this way, the current sensitivity and thermal sensitivity are included in the subsequent mismatch identification without changing the device physical model itself.

[0027] Please refer to Figure 2 , Figure 2 The horizontal axis represents the sorting of response edges, based on their coupling coefficients. The results are arranged from high to low, with the vertical axis representing the dimensionless comprehensive response value. Figure 2 Corresponding to the aforementioned coupling coefficient calculation relationship , Figure 2 Each bar in the diagram represents a response edge from an interface sub-region to the internal conductive region. The data includes both the current response caused by changes in potential boundaries and the heat source response caused by changes in temperature boundaries. The coupling coefficient in the legend represents the final ranking index after weighted synthesis of the two types of responses, rather than individual current or heat source values.

[0028] from Figure 2 As can be seen, the top-ranked response edges are concentrated in positions such as RE-080, RE-035, RE-025, RE-070, and RE-026, with RE-080 corresponding to... The value is approximately 7.071 for RE-035 and approximately 6.971 for RE-035. The first five digits are all higher than 6.47, indicating that these interface sub-regions are most sensitive to boundary disturbances compared to their internal conductive regions. (Combined with the symbol...) , , It can be seen that the larger current-voltage response terms are the main reason for the increase in the front column values; the electrical response of the front response edge accounts for about 96.91% to 98.99% in the derived data, indicating that the ranking results in this group of samples are mainly driven by current migration sensitivity.

[0029] The thermal response term still has a corrective effect on the ranking. (Symbol) , and The normalized response of the corresponding heat source power as the temperature boundary changes, although the proportion of the first thermal response is about 1.01% to 3.09%, which is lower than that of the electrical response, it will affect the order of multiple similar response edges.

[0030] In a preferred embodiment, the packaged conductive network model defines the chip mounting area and external power terminals as network nodes, the continuous conductors connecting the network nodes as network branches, and establishes a pose mapping for each chip instance that transforms the chip local coordinates to the package layout coordinates. The pose mapping projects the interface nodes in the distributed conduction response descriptor to the corresponding chip mounting area, establishes the interface node-network branch association according to the geometric coverage relationship between the projected interface boundary and the package contact boundary, and merges the associations that share the same package contact boundary into the same current continuity constraint.

[0031] In this embodiment, network nodes do not merely represent geometric points in the layout, but rather geometric regions with approximately equipotential relationships or continuous thermal boundary relationships. Chip mounting area nodes store the chip back contact area and mounting layer area, while external power terminal nodes store the terminal surfaces connected to external circuits. Network branches are generated from continuous conductors between two network nodes. During generation, a central path is established along the conductor's length, and the conductor cross-section, material region, and surface heat dissipation boundary are associated with the central path. Pose mapping consists of translation and rotation relationships. The system transforms each vertex of the interface polygon in the chip's local coordinates to the package's global coordinates, and then... The encapsulation contact polygons within the particle mounting area undergo intersection operations to obtain the encapsulation mesh actually covered by the interface boundary. The interface node-network branch association saves the overlapping mesh index instead of just the branch number. Therefore, the same interface node can connect to one or more network branches based on the contact coverage, and the same network branch can also receive current returned from multiple interface nodes. The system establishes current continuity constraints at the common encapsulation contact boundary, summarizes the current flowing into each interface node in a unified positive direction, and uses the summary result as the boundary input of the corresponding network branch. After the encapsulation solution is completed, the node potential is returned to each interface node according to the overlapping mesh.

[0032] Furthermore, the pose mapping and descriptor version are kept bound together. When the position of a chip instance is swapped or its orientation changes, the system does not modify the device mesh inside the power MOS chip model. Instead, it recalculates the projection contours and overlapping meshes of the interface nodes in the package global coordinates. The original interface node identifiers, internal region node identifiers, and response edge identifiers remain unchanged. The network branch association generates a new mapping version. Each mapping version saves the chip instance identifier, the coordinate relationship before and after the transformation, the package contact boundary reference, and the overlapping mesh set. When the package model calls the mapping version, it checks whether the chip instance identifier and the descriptor version are consistent. If the chip structure has not changed but the package layout has changed, the original descriptor is reused and the network branch association is updated. If the chip structure change causes changes in the electrode contours or internal region nodes, a new descriptor version is generated synchronously and reprojected. The current continuity constraint is automatically reconstructed based on the current mapping version, so that layout candidates can be compared on the basis of the same device response, and device structure candidates can also be compared on the basis of the same package network.

[0033] In a preferred embodiment, the power MOS chip model and the packaged conductive network model are coupled according to an alternating solution sequence. The packaged conductive network model first forms the potential and temperature boundaries of each network node based on the existing interface sub-region current and writes them into the corresponding interface node. The power MOS chip model then updates the distributed conduction response descriptor based on the written boundaries and returns the new interface sub-region current and local loss. Each round of coupling saves the mapping version between the interface node and the network branch, and the consistency of the branch current order and local loss position in adjacent mapping versions is used as the coupling end marker.

[0034] In this embodiment, the alternating solution starts from the baseline boundary state of the power MOS chip model. The device solver writes the current of each interface sub-region and the local loss of each internal region into the shared data area. The shared data area merges the current into the network branch corresponding to the current mapping version according to the chip instance identifier and interface node identifier, and projects the local loss onto the heat source grid of the chip mounting area. The package electrical solver calculates the network node potential and branch potential drop under the condition that the external power terminal boundary is fixed. The package thermal solver calculates the temperature distribution of the chip mounting area based on the device local loss, conductor loss and thermal boundary. The system then reads the potential and temperature from the overlapping grid associated with the interface node and converts them to the chip local coordinates. The device solver uses the return boundary to resolve the carrier distribution and heat source distribution. The updated interface current, local loss and response edge record form the next round of shared data. The system assigns a continuous version identifier to the results of each round and saves the references between the input boundary version, device result version, package result version and mapping version, so that when an abnormal result occurs in any round, the source of the abnormality can be located along the version reference without mixing data from other rounds.

[0035] Specifically, the coupling termination determination does not only compare the overall on-resistance, but also performs discrete comparisons of the branch current order and the location of local losses. The system arranges network branches with common source and drain network nodes according to their current response, and projects the local loss regions in the device mesh onto the internal region nodes and establishes labels according to their spatial locations. When the arrangement of branch currents remains unchanged in two adjacent rounds and the highest local loss position belongs to the same internal region node, the current mapping version is recorded as a stable version. If the arrangement or position label changes, the current package boundary is written back to the device model for the next round of solution. When multiple branches with the same arrangement position appear, the system retains the parallel relationship and does not replace them with arbitrary branches. When local loss crosses the internal region boundary, the system recalculates the loss attribution based on the affected mesh and then compares them. After the stable version is generated, the results of the previous rounds are still saved so that the previous boundary can be called as a new initial state after the chip pose or package contact boundary is adjusted. The potential direction, temperature reference, and positive current direction used in the alternating solution remain consistent in all versions.

[0036] In a preferred embodiment, before writing the device solution results and the package solution results into the shared data area, dimensional and mesh consistency processing is performed. The system reads the potential, current, power, and temperature dimensions declared in each solution domain, converts the values ​​to a unified internal dimension, and then associates the interface nodes. When the device surface mesh and the package contact mesh do not coincide, the overlapping polygon is used as an intermediate projection plane. The current flux on the device mesh is aggregated to the package mesh according to the projection plane coverage relationship, and the potential and temperature on the package mesh are written back to the device boundary according to the same projection plane. The total current before and after projection is calculated under the parent identifier of the interface node. When the two are inconsistent, the current coupling cycle is stopped and the overlapping polygons are subdivided again. The local loss projection is distributed to the heat source mesh of the core mounting area using the power conservation method. When there is a situation where the interface node lacks encapsulation contact, the network branch is not connected to the external terminal, or the internal conduction chain does not reach the drain contact boundary, the system writes the corresponding identifier into the anomaly record and prohibits it from entering the stable version judgment. After the solver is interrupted, it reloads the most recent mapping version with complete boundary and current continuity. Only the interface node, network branch and internal conduction chain pointed to by the anomaly record are recalculated to keep the numerical anomaly handling and version relationship consistent.

[0037] In a preferred embodiment, the internal conductive region is traced along the source-drain current direction by the device mesh, and the region boundary is established at the position where the current direction bifurcates or the potential gradient contribution is transferred. The internal region nodes that start from the same interface node and are continuously connected to the drain contact boundary are connected in sequence to form an internal conductive chain. The response edge records the current migration record and heat source migration record of the boundary state change in the corresponding internal conductive chain. Each internal conductive chain is associated with the corresponding package branch through the interface node index.

[0038] In this embodiment, path tracing starts with the device surface mesh covered by the interface node. For each starting mesh, the current vector is read and the subsequent mesh pointing to the drain contact boundary is determined. The selection of the subsequent mesh is based on the current continuity direction and material connection relationship at the shared mesh surface. When there are multiple subsequent meshes in the same mesh, a branching mark is established and the meshes after the branching are included in different candidate paths. When the candidate paths re-merge, a shared internal region node is established without copying the device mesh. The potential gradient contribution is determined based on the potential change position of adjacent material regions in the same current direction. When the main potential change in the continuous mesh is transferred from one material region to another, a region boundary is established at the transfer interface. Thus, the continuous mesh extending from below the source metal through the adjacent channel region, drift region and substrate region to the drain contact boundary is organized into a directed internal conduction chain. Each internal conduction chain saves the starting interface node, the internal region nodes passed through, the terminating drain interface node and the corresponding response edge. When the path tracing encounters a closed loop, the loop branch that cannot reach the drain contact boundary is deleted according to the main current direction of the source and drain. When an isolated mesh with approximately zero current is encountered, it is kept in the device model but not written into the current conduction chain.

[0039] Preferably, the system performs integrity checks on the internal conduction chains. The checks include whether the starting interface node belongs to the source electrode, whether the ending interface node belongs to the drain electrode, whether adjacent internal region nodes share the device mesh boundary, and whether the response edge direction is consistent with the net current direction in the benchmark solution. When any internal region node has multiple incoming or outgoing edges, the system saves the forking relationship based on the current distribution record on the shared boundary and does not forcibly merge multiple chains into a centralized channel. After the boundary state changes, the system re-accumulates the current changes and heat source changes of each internal region along the original internal conduction chain and reassigns the meshes whose changes exceed the migration relationship of adjacent regions to the corresponding internal region nodes. If the reassignment does not change the node connection relationship, only the response edge record is updated. If the reassignment causes the original chain to break or generates a new fork, the node connection is rebuilt in the affected chain. The interface node index remains fixed before and after reconstruction, so that the package branch association continues to point to the same electrode position. The data structure of the internal conduction chain also saves the device mesh version and the boundary state version to prevent regional nodes of different process structures or different coupling cycles from being combined into the same chain.

[0040] In a preferred embodiment, after the interface node is projected onto the package contact boundary, it is divided into covered sub-nodes and non-covered sub-nodes according to the overlapping grid of the projected boundary and the package contact boundary. Only the covered sub-nodes are connected to the corresponding network branches. The covered sub-nodes receive the package potential boundary according to the index relationship between the overlapping grid and the internal conductive chain. The temperature boundary formed by the chip mounting layer is written into the internal region node according to the same overlapping grid. The returned interface sub-region current flows into the network branch along the package contact boundary corresponding to the covered sub-node.

[0041] In this embodiment, the projected boundaries of the interface nodes and the package contact boundaries are both converted into overlay sets on the same package surface mesh. The system performs a mesh intersection operation on the two overlay sets. Mesh sets within the intersection form overlay child nodes, while mesh sets within the interface projection set but not within the intersection form non-overlay child nodes. The original interface node identifier is retained as the parent identifier, and overlay and non-overlay child nodes obtain child identifiers respectively. Overlay child nodes connect to the corresponding network branches based on the continuous conductors where the overlapping meshes are located. Non-overlay child nodes do not receive package potential boundaries and maintain internal device continuity along the die surface electrode model. When an interface projection... When the boundary intersects with multiple discontinuous package contact areas, the system establishes an independent cover sub-node for each continuous intersection. Each cover sub-node receives the potential result of its respective network branch. The temperature boundary is written into the corresponding device mesh through the thermal projection relationship between the chip mounting layer and the chip surface. The potential projection and temperature projection use the same overlapping mesh index but save the boundary type separately. The interface sub-region current returned by the device model is first aggregated according to the current flux on the mesh corresponding to the cover sub-node, and then written into the network branch connected to the cover sub-node. The current of the cover sub-nodes sharing the same package contact area is aggregated according to the current continuity constraint.

[0042] Furthermore, the equivalent on-resistance of the cross-scale conduction chain after packaging can be quantified by the internal resistance of the core, the resistance of the packaged branch, and the resistance increment caused by temperature rise, and the calculation relationship is as follows.

[0043] ; in, This represents the equivalent on-resistance of multiple parallel cross-scale conduction chains after packaging, where n represents the number of cross-scale conduction chains connecting the common source network node and the common drain network node. This represents the core internal resistance obtained from the device mesh in the k-th cross-scale conduction chain. This represents the package resistance of network branches connected by the same cross-scale conduction link. This indicates the temperature coefficient of the on-resistance of the power chip corresponding to this chain. This indicates the temperature rise of the chain relative to the device reference temperature. This represents the resistance increment caused by temperature rise. In the example simulation results, there are two parallel cross-scale conduction chains. The first chain has an internal core resistance of 2 milliohms, a package resistance of 0.4 milliohms, and a resistance increment of 0.2 milliohms due to temperature rise. The second chain has an internal core resistance of 2.2 milliohms, a package resistance of 0.3 milliohms, and a resistance increment of 0.1 milliohms due to temperature rise. After substituting these values, the total branch resistance of both chains is 2.6 milliohms, and the equivalent conduction resistance after packaging is 1.3 milliohms. If the layout mapping changes the package resistance of the first chain to 0.8 milliohms and the resistance increment of temperature rise to 0.4 milliohms, while keeping the second chain the original simulation result, then the equivalent conduction resistance after packaging becomes approximately 1.43 milliohms. The values ​​here are used as a calculation example of coupling results rather than hardware structure settings to illustrate that interface coverage relationships and thermal boundaries must be included in the candidate layout for comparison along with the internal core resistance.

[0044] Please refer to Figure 3 , Figure 3 The horizontal axis represents the number of rounds in which the device model and the packaged conductive network model are solved alternately, ranging from round 1 to round 10. The vertical axis represents the equivalent on-resistance after packaging, in milliohms. The solid black line represents the calculation results of each round. The gray dashed line and gray dotted line serve as references for the target level and warning level, respectively. Figure 3 Corresponding to the aforementioned equivalent on-resistance calculation relationship The curve value is determined by the resistance of multiple parallel cross-scale conductive chains, rather than by the resistance of any single chain alone.

[0045] exist Figure 3 In the corresponding data, the number of conducting chains, n, participating in the parallel calculation is 24. The total branch contribution of each chain is... , as well as Together they form, among which Reflects the internal conduction path of the core particle. Reflects the resistance of the packaged branch. It reflects the increase in resistance caused by the temperature rise. Figure 3 Round 1 The value was approximately 0.102229 milliohms in the first round and approximately 0.102223 milliohms in the 10th round, with an overall change of approximately 0.000006 milliohms. The curve showed a slight decrease and tended to stabilize.

[0046] Looking at the changes in rounds, the current sequence stability identification from round 1 to round 7 has not yet been completed, and the local loss area is still mainly IR-016; after round 8, the local high loss area turns to IR-024 and remains stable; in round 10, both current sequence stability and loss area stability are satisfied.

[0047] In a preferred embodiment, after the cross-scale conduction chain is solved, the internal resistance source, network branch source, and temperature boundary source are saved. The internal resistance of the device is extracted from the device grid potential drop and current response through which the internal conduction chain passes. The package branch resistance is extracted from the conductor geometry and material conductivity results of the network branch. The temperature rise resistance increment is generated by associating the temperature boundary returned from the core mounting area with the temperature-related conduction results in the device model. The system does not pre-merge the three types of resistance sources into an indivisible lumped value, but retains their respective data references when comparing candidate layouts. When the equivalent resistance of a cross-scale conduction chain changes, the change can be determined from the core internal response, package contact coverage, or local temperature boundary by following the reference. The splitting of the coverage sub-nodes also allows the potential drop of each branch to be calculated separately when multiple package branches are connected to the same interface sub-region. The surface area corresponding to the non-coverage sub-nodes continues to participate in the device-side current extension calculation but does not generate fictitious package current input, thereby maintaining the consistency between the device grid boundary and the actual package contact range.

[0048] In a preferred embodiment, please refer to Figure 4 After each round of coupling, each internal conduction chain and its associated network branches are connected in series to form a cross-scale conduction chain. Under the same terminal perturbation, the current response order of the core and the potential drop contribution order of the package branch are formed respectively. Cross-scale conduction chains with common source network nodes and common drain network nodes are compared. The internal conduction chain with the higher current response order is connected to the network branch with the same higher potential drop contribution order and the association is marked as a conduction mismatch edge. A mismatch topology graph is formed based on the core instance or network branch shared by the conduction mismatch edge.

[0049] In this embodiment, the concatenation of cross-scale conduction chains is completed based on the continuous index between interface nodes, cover sub-nodes, and network branches. The system reads the internal conduction chain starting from the source interface node, enters the encapsulation network through the cover sub-node, and reaches the common source network node along the network branch. Simultaneously, it reaches the common drain network node from the drain interface node through the corresponding cover sub-node and encapsulation branch. Source-drain paths belonging to the same power chip and connected to the same pair of external network nodes are merged into a single cross-scale conduction chain record. Unified terminal perturbation is applied to all candidate chains using the same external terminal boundary change. On the component side, the current response order is formed from high to low according to the current change of each internal conduction chain. On the package side, the potential drop contribution of each network branch under the unit current condition is arranged from high to low. The system compares the two orders with the interface node-network branch association as the corresponding key. If the internal conduction chain with the higher current response position is connected to the network branch with the same higher potential drop contribution position, a conduction mismatch edge is generated. The conduction mismatch edge saves the internal conduction chain identifier, the overriding child node identifier, the network branch identifier, and the two contribution positions. When the positions are tied, the tied set is saved and processed together in the subsequent layout candidates.

[0050] Specifically, the mismatch topology graph uses core instances and network branches as topology nodes, and conductive mismatch edges as the connecting edges between nodes. When the same core instance contains multiple conductive mismatch edges, the system saves the distribution of these edges on the core electrode surface according to the local coordinate positions of the interface nodes. When the same network branch connects multiple core instances and forms multiple conductive mismatch edges, the system marks the network branch as a shared mismatch branch. The mismatch topology graph also retains normal associations that are not marked as mismatches as adjacency references to distinguish between local interface mismatches and shared branch mismatches. Instead of directly replacing the potential drop contribution with local temperature variations, the system reflects the temperature in the cross-scale conduction chain data through the current response and resistance increment formed by the device model. The system generates an independent mismatch topology graph for each stable mapping version and associates it with the corresponding package layout through the mapping version identifier. If the subsequent pose exchange only changes some chip instances, only the topology nodes and associated edges connected to these instances are updated. Unaffected normal associations and conduction mismatch edges continue to be reused. The topology graph generation result serves as the input for package layout reconstruction without directly modifying the device physical structure.

[0051] In a preferred embodiment, the conducting mismatched edges in the mismatched topology diagram are used as the adjustment objects. While keeping the power MOS chip model itself unchanged, the pose mapping is first exchanged between chip instances with the same electrode definition, and the interface node-network branch association is migrated synchronously with the pose mapping. When the exchanged association still retains the original conducting mismatched edges, the package contact boundary is reconstructed along the adjacent network branches connecting the conducting mismatched edges in the mismatched topology diagram. The reconstruction result is rewritten into the package conductive network model, and only the candidate package layouts of the conducting mismatched edges with higher potential drop contribution positions without new additions are retained.

[0052] In this embodiment, pose swapping is performed only between chip instances with compatible electrode definitions and consistent device model versions. The system reads the chip instance identifier and network branch identifier at both ends of the mismatched conduction edge, searches for chip instances in the candidate mounting area that maintain semantic consistency of source, drain, and gate connections, and then swaps the corresponding pose mapping without copying or modifying the chip device mesh. After the swap, the interface projection boundary, covered sub-nodes, and network branch associations are recalculated, and the existing distributed conduction response descriptor is called to perform alternating solution. If the original mismatched conduction edge disappears and no mismatched edge with a higher potential drop contribution ranking is generated in other interface nodes, the pose swap candidate is saved. If the original mismatched conduction edge still exists, the system reads the adjacent branches of the network nodes shared with the mismatched edge along the mismatched topology graph, expands or shrinks the package contact boundary according to the interface projection coverage, and keeps the contact boundary within the corresponding continuous conductor range. The reconstructed package contact boundary is re-discretized into a package mesh, and after performing an intersection operation with the chip interface projection, it is written into a new mapping version, and then the device model and package model are coupled and verified.

[0053] Preferably, the candidate package layout is screened using a set of mismatched edge positions. The system saves the potential drop contribution position of all current conducting mismatched edges for each candidate and aligns it edge by edge with the mismatched topology before adjustment. If a candidate deletes a conducting mismatched edge with a higher position in the original position but adds a conducting mismatched edge with a higher position in other chip instances or shared network branches, the candidate is discarded. If a candidate only retains mismatched edges with lower positions or converts the original mismatched edges into normal associations, the next round of alternating solution is entered. When multiple candidates have the same set of mismatched edge positions, the system continues to compare the equivalent on-resistance and local loss positions of the packaged cross-scale conducting chains, but does not cover the mismatched topology differences with a single overall value. The candidate screening process saves the parent-child relationship between each pose exchange, contact boundary reconstruction, mapping version change and mismatched graph change. If a subsequent candidate fails to converge in device solution, the package branch is disconnected, or the current continuity constraint is not met, the system reverts to the corresponding parent version and excludes the abnormal layout transformation, keeping the candidate generation process verifiable.

[0054] Furthermore, the local update of the mismatched topology graph adopts the affected domain propagation method. After the pose exchange occurs, the system writes the chip instance participating in the exchange, the overlay child node associated with its interface node, and the network branches connecting the overlay child node into the initial affected domain. Then, it accesses adjacent branches that share the potential boundary or thermal boundary with the initial affected domain along the network node connection relationship. Branches that may receive redistributed current or heat sources are added to the extended affected domain. Access terminates at independent network nodes where the external power terminal or boundary conditions have not changed. Alternate solution updates the package mesh and corresponding device mesh only within the affected domain, and continues outside the affected domain. Using parent version data, the new mismatch topology graph copies the topology nodes and associated edges that are not connected to the affected domain in the parent version, and recalculates the contribution order and conduction mismatch edges of the cross-scale conduction chains within the affected domain. When candidates have the same chip pose mapping, the same package contact boundary, and the same set of mismatch edges, the system determines them as duplicate candidates based on the version summary of the three. Only records that have completed stable mapping determination and have shorter parent version chains are retained. The device results and package results associated with duplicate candidates are saved as shared references and are not written repeatedly. This allows each branch in the candidate tree to be traced back to the pose exchange or contact boundary reconstruction operation that generated the branch.

[0055] In a preferred embodiment, the power MOS chip model also receives a semiconductor process simulation structure corresponding to the candidate package layout, and projects the metal contact contours and doped boundaries in the process simulation structure to the interface nodes and internal region nodes; when the node connection relationship before and after projection is consistent, the current migration record and heat source migration record of the corresponding response edge are updated with the projected device mesh; when the node connection relationship changes, only the internal conduction chain crossed by the changed boundary is reconstructed, and the coupling and mismatch topology graph generation is re-executed on the updated distributed conduction response descriptor.

[0056] In this embodiment, the semiconductor process simulation structure and the power MOS chip model are associated through unified device coordinates and material region identifiers. The system extracts the metal contact contour, junction contour, and doping boundary from the process structure and projects the extraction results onto the current device mesh. When the change in the metal contact contour causes a change in the interface projection range, the system redetermines the boundary surface set of the affected interface sub-region. When the doping boundary changes and crosses the internal conduction region, the system marks the device mesh that is crossed and its internal conduction chain. The interface nodes, internal region nodes, and response edges before and after projection are aligned according to the geometric overlap and path continuity relationship. If the node connection relationship remains consistent, it means that the original interface node can still reach the drain contact boundary along the same internal region node order. At this time, the directed graph topology is reused and the current migration record and heat source migration record in the response edge are updated using the solution results of the new device mesh. If the node connection relationship changes, it means that a new branch appears in the conduction path, the original branch disappears, or the region boundary crosses the original internal node. At this time, only the internal conduction chain segment crossed by the changed boundary is deleted, and the unaffected nodes are retained at both ends of the changed segment and local path tracing is re-executed.

[0057] Furthermore, after local path tracing is completed, the system checks the shared mesh boundary and current direction between newly created internal region nodes and retained nodes. When the connection is established, the new chain segment is embedded back into the original internal conduction chain. When the connection is not established, the reconstruction range is expanded until an interface node or drain contact boundary with a stable connection relationship is reached. The updated descriptor obtains a new device structure version and inherits the historical record of the unchanged response edges. The package conductive network model continues to use the original candidate layout, but updates the covered and non-covered child nodes according to the new interface projection contour. Then, the alternating solution is re-executed and the corresponding mismatch topology graph is generated. The system saves the mapping versions of the same candidate package layout under different process structures in parallel. The version record includes the process structure source, the reconstructed internal conduction chain set, the reused response edge set, and the newly generated mismatch edge set. Device meshes, internal conduction chains, and package branches that are not affected by the changed boundaries do not participate in the repeated construction. They only receive new boundary results in the coupled solution, so that the process structure changes can be transmitted along the existing model relationship without interrupting the comparison basis between candidate layouts.

[0058] In a preferred embodiment, for each mismatch topology diagram formed by simulation structures of different semiconductor processes, the candidate package layout is projected into a cross-process structure candidate set using the chip instance identifier and the interface node-network branch association as the alignment benchmark. The cross-process structure candidate set retains candidates in each mismatch topology diagram where the same conduction mismatch edge does not appear repeatedly, and the version relationship of the chip pose mapping, package contact boundary, and distributed conduction response descriptor corresponding to the retained candidates is encapsulated into low on-resistance MOSFET system design data.

[0059] In this embodiment, the cross-process structure candidate set uses the candidate package layout identifier as the main index. The mismatch topology generated under each process structure is used as a sub-record under the main index. The system uses the chip instance identifier to align the power chips in different sub-records, and then uses the local coordinates of the interface node and the network branch layout reference to align the interface node-network branch association. When the process structure changes cause the interface node to be split or merged, the system establishes a many-to-one or one-to-many alignment relationship through the parent interface node identifier and geometric coverage, and classifies the conduction mismatch edges generated by the same original interface range into the same mismatch edge family. If the same conduction mismatch edge appears repeatedly in multiple process structures, it means that the mismatch between the current response inside the corresponding chip and the potential drop contribution of the package branch does not disappear with the change of process structure. The system removes the candidate from the cross-process structure candidate set. If the same mismatch edge does not appear repeatedly in each process structure, the candidate is retained, and the system continues to check the current continuity constraints, mapping version integrity, and whether the cross-scale conduction chain connects to the common source-drain network node in each sub-record. After the check is passed, a candidate retention record is generated.

[0060] In an extended embodiment, the external circuit models of candidate batteries connected to the on / off state, motor arm, or synchronous rectifier power stage are retained. The external circuit models only provide the terminal potential boundaries and load current boundaries within the conduction interval to the packaged conductive network model, without changing the power MOS chip model and interface node definitions. The system extracts boundary snapshots corresponding to different conduction states from the external circuit operation sequence. For each boundary snapshot, the system calls the packaged chip pose mapping, package contact boundary, and distributed conduction response descriptor to perform alternating solutions. Then, the resulting cross-scale conduction chains are grouped according to common source-drain network nodes, and the same candidate is checked in each... Whether duplicate conduction mismatch edges appear under boundary snapshots, when a boundary snapshot causes the local loss location to migrate, the system determines the internal region node after migration along the internal conduction chain index, and returns the corresponding temperature boundary to the device mesh to continue solving. When an external state causes the original network branch to exit conduction, only the chain associated with the branch is removed from the current cross-scale conduction chain set, without deleting the mapping version and descriptor version. After all boundary snapshots are completed, the branch current order, local loss location, mismatch topology map and equivalent on-resistance after packaging corresponding to each state are written into the candidate retention record to form the verification data for the connection relationship of specific power stage.

[0061] Specifically, the design data for low on-resistance MOSFET systems adopts a traceable data packet structure. The data packet retains candidate identifiers, connecting chip pose mapping versions, package contact boundary versions, and distributed conduction response descriptor versions. It also saves the corresponding device mesh references, internal conduction chain references, overlay child node references, network branch references, coupling result references, and mismatch topology references for these versions. The layout output records the position and orientation of chip instances in the package global coordinates, interface projection boundaries, and network branch geometry. The model output records the response edges of each interface node, the resistance source of each cross-scale conduction chain, and the temperature boundary source. The verification section records the candidate alignment relationships under different process structures and the mismatch edge families of removed candidates. When the design data is reloaded, the system restores the power MOSFET chip model and package conductive network model according to the version relationship, re-executes the alternating solution of the stable mapping version, and compares the restored branch current order, local loss location, and mismatch topology with the records in the data packet for consistency. If they match, the design data is encapsulated. If they do not match, the system locates the changed device structure, interface mapping, or package branch along the version reference and regenerates the affected records.

Claims

1. A low on-resistance MOSFET system based on a core particle technology, characterized in that, It includes a power MOS chip model and a packaged conductive network model coupled to the power MOS chip model. The power MOS chip model forms interface sub-regions according to the chip electrode regions, and generates distributed conduction response descriptors based on the current distribution and heat source distribution of each interface sub-region under the potential boundary and temperature boundary. The packaged conductive network model maps the distributed conduction response descriptors to the corresponding package branches according to the chip arrangement position and electrode connection relationship. It feeds back the interface sub-region boundary to the power MOS chip model and receives the returned interface sub-region current and local loss, so as to determine the chip combination and package interconnection scheme based on the feedback conduction response.

2. The low on-resistance MOSFET system based on chip technology according to claim 1, characterized in that, The power MOS chip model discretizes the source metal projection boundary and the drain contact projection boundary into the interface sub-regions, and merges the grid regions in the device grid that are located on the same continuous conduction path into the internal conduction region. The distributed conduction response descriptor adopts a directed graph. The interface node in the directed graph is connected to the corresponding internal region node. The response edge connecting the two is associated with the current migration record and heat source migration record before and after the boundary state change of the interface sub-region, and the index relationship between the interface node and the geometric position of the core electrode is retained.

3. The low on-resistance MOSFET system based on chip technology according to claim 1, characterized in that, The packaged conductive network model defines the chip mounting area and external power terminals as network nodes, the continuous conductors connecting the network nodes as network branches, and establishes a pose mapping from chip local coordinates to package layout coordinates for each chip instance. The pose mapping projects the interface nodes in the distributed conduction response descriptor to the corresponding chip mounting area, establishes the interface node-network branch association according to the geometric coverage relationship between the projected interface boundary and the package contact boundary, and merges the association relationship sharing the same package contact boundary into the same current continuity constraint.

4. The low on-resistance MOSFET system based on chip technology according to claim 1, characterized in that, The power MOS chip model and the packaged conductive network model are coupled according to an alternating solution sequence. The packaged conductive network model first forms the potential and temperature boundaries of each network node based on the existing interface sub-region current and writes them into the corresponding interface node. The power MOS chip model then updates the distributed conduction response descriptor based on the written boundaries and returns the new interface sub-region current and local loss. Each round of coupling saves the mapping version between the interface node and the network branch, and the coupling ends when the branch current order and local loss position in adjacent mapping versions remain consistent.

5. The low on-resistance MOSFET system based on chip technology according to claim 2, characterized in that, The internal conductive region is traced along the source and drain current direction by the device grid, and the region boundary is established at the position where the current direction bifurcates or the potential gradient contribution is transferred. Internal region nodes that start from the same interface node and are continuously connected to the drain contact boundary are sequentially connected to form an internal conduction chain. The response edge records the current migration record and heat source migration record of the boundary state change in the corresponding internal conduction chain. Each internal conduction chain is associated with the corresponding package branch through the interface node index.

6. The low on-resistance MOSFET system based on chip technology according to claim 3, characterized in that, After the interface node is projected onto the encapsulation contact boundary, it is divided into covered sub-nodes and non-covered sub-nodes according to the overlapping grid of the projected boundary and the encapsulation contact boundary, and only the covered sub-nodes are connected to the corresponding network branches. The covered sub-nodes receive the package potential boundary according to the index relationship between the overlapping grid and the internal conductive chain. The temperature boundary formed by the chip mounting layer is written into the internal region node according to the same overlapping grid. The returned interface sub-region current flows into the network branch along the package contact boundary corresponding to the covered sub-node.

7. The low on-resistance MOSFET system based on chip technology according to claim 1, characterized in that, After each round of coupling, each internal conduction chain and its associated network branches are connected in series to form a cross-scale conduction chain, and under the uniform terminal perturbation, the order of current response inside the core and the order of contribution of unit current potential drop of the packaged branch are formed respectively. The cross-scale conduction chains with common source network nodes and common drain network nodes are compared accordingly. The internal conduction chains with higher current response rankings are connected to network branches with higher potential drop contribution rankings and the association is marked as a conduction mismatch edge. A mismatch topology graph is formed based on the core instance or network branch shared by the conduction mismatch edge.

8. The low on-resistance MOSFET system based on chip technology according to claim 7, characterized in that, Taking the mismatched conduction edges in the mismatched topology diagram as the adjustment object, while keeping the power MOS chip model itself unchanged, first exchange the pose mapping between chip instances with the same electrode definition, and then migrate the interface node-network branch association synchronously with the pose mapping. When the original mismatched conduction edge is still retained after the exchange, the package contact boundary is reconstructed along the adjacent network branches connecting the mismatched conduction edge in the mismatched topology graph. The reconstruction result is rewritten into the package conductive network model, and only the candidate package layouts of the mismatched conduction edge with the highest potential drop contribution position are retained.

9. The low on-resistance MOSFET system based on chip technology according to claim 8, characterized in that, The power MOS chip model also receives a semiconductor process simulation structure corresponding to the candidate package layout, and projects the metal contact contour and doping boundary in the process simulation structure to the interface node and internal region node. When the node connection relationship before and after projection is consistent, the current migration record and heat source migration record of the corresponding response edge are updated with the device mesh after projection. When the node connection relationship changes, only the internal conduction chain crossed by the changed boundary is reconstructed, and coupling and the generation of the mismatched topology graph are re-executed on the updated distributed conduction response descriptor.

10. The low on-resistance MOSFET system based on chip technology according to claim 8, characterized in that, For each mismatched topology diagram formed by simulation structures of different semiconductor processes, the candidate package layout is projected into a cross-process structure candidate set, using the chip instance identifier and the interface node-network branch association as the alignment benchmark. The cross-process structure candidate set retains candidates whose conduction mismatch edges do not appear repeatedly in each mismatch topology diagram, and encapsulates the chip pose mapping, package contact boundary, and version relationship of the distributed conduction response descriptor corresponding to the retained candidates into low on-resistance MOSFET system design data.