Search

CN-120724964-B - Reinforced learning-based GDS2/OASIS layout DRC violation repair method

CN120724964BCN 120724964 BCN120724964 BCN 120724964BCN-120724964-B

Abstract

The invention discloses a reinforced learning-based GDS2/OASIS layout DRC violation repair method, which relates to the technical field of Electronic Design Automation (EDA) and integrated circuit physical layout repair, and comprises the steps of reading GDS2/OASIS layout files and extracting a graphic set of a target layer; calculating DRC violation number in a GDS2/OASIS layout through a DRC detection module to obtain a region to be repaired in the GDS2/OASIS layout, converting a graph set of the region to be repaired in the GDS2/OASIS layout into a state representation of the layout, constructing a repair strategy module for generating a graph adjustment action according to the state representation of the layout, and applying the graph adjustment action to a target graph in the GDS2/OASIS layout. The invention realizes automatic DRC repair of GDS2/OASIS format layout, has stronger universality, precision and expandability, and can effectively promote the automation and intellectualization level of back-end design.

Inventors

  • WU CHAOHUI
  • HUANG SONGTING
  • LI BIN
  • HUANG WENLI
  • Qin Chaozheng

Assignees

  • 华南理工大学

Dates

Publication Date
20260512
Application Date
20250605

Claims (7)

  1. 1. A reinforcement learning-based GDS2/OASIS layout DRC violation repair method is characterized by comprising the following steps: Reading GDS2/OASIS layout files, and extracting a graph set of a target layer; calculating DRC violation number in a GDS2/OASIS layout through a DRC detection module to obtain a region to be repaired in the GDS2/OASIS layout; Converting a graph set of the region to be repaired in the GDS2/OASIS layout into a state representation of the layout; Constructing a repair strategy module for generating a graph adjustment action according to the state representation of the layout; applying the graph adjustment action to a target graph in the GDS2/OASIS layout; the repair strategy module comprises the following steps: Firstly, extracting a current layout state s t from a state representation of the layout; Step two, defining an action space A, extracting an action a t from the action space A according to the current layout state s t , and executing the action a t to obtain a new state s t+1 ; Wherein, the action space A is expressed as: ; Wherein, the Representing the direction of the ith rectangular metal pattern P i A translation distance delta; A sub-region R j in the jth complex polygonal metal graph is represented for local translation, wherein the sub-region R j represents a rectangular segment marked as illegal by the DRC detection module; Indicating that the size of the designated side of the ith rectangular metal pattern P i is adjusted, increased or decreased ; The designated sides of the sub-rectangular region R j of the jth complex polygonal metal pattern are shown as being resized, noOp is shown as keeping the current pattern unchanged; thirdly, calling the DRC detection module to obtain the DRC violation number after executing the action a t ; Fourthly, calculating rewards r t according to the current layout state s t , the new state s t+1 and the DRC violation number after the action a t is executed; wherein the prize r t is calculated by the following formula: ; Wherein E (s t ) represents the total number of DRC errors in the current state; Fifthly, constructing a sample (s t ,a t ,r t ,s t+1 ) according to the current layout state s t , the new state s t+1 , the execution action a t and the rewards r t , storing the sample (s t ,a t ,r t ,s t+1 ) into an experience pool, and enabling the sample to participate in subsequent training; and sixthly, repeating the first step to the fifth step until the DRC violation number is stabilized below a set threshold or the repair strategy converges or reaches the upper limit of the training period.
  2. 2. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 1, wherein in a second step, after performing the action a t , the new state s t+1 is represented as: ; Where T is a state transfer function.
  3. 3. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 1, wherein in the sixth step, the repair policy convergence is specifically: the amplitude of the ripple of the prize r t is below a preset range.
  4. 4. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 3, wherein the preset range is +/-1% to +/-2%.
  5. 5. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 1, wherein in the sixth step, the set threshold is 3-10.
  6. 6. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 1, wherein the graphic adjustment action is applied to a target graphic in a GDS2/OASIS layout through a KLayout Python API or gdspy library.
  7. 7. The reinforcement learning-based GDS2/OASIS layout DRC violation repair method according to claim 1, wherein if the GDS2/OASIS layout modified by the graphic adjustment action still has DRC violations, the GDS2/OASIS layout file modified by the graphic adjustment action is continuously read and violation repair is executed on the GDS2/OASIS layout file.

Description

Reinforced learning-based GDS2/OASIS layout DRC violation repair method Technical Field The invention relates to the technical field of Electronic Design Automation (EDA) and integrated circuit physical layout repair, in particular to a GDS2/OASIS layout DRC violation repair method based on reinforcement learning. Background In the current field of integrated circuit design, error correction for circuit layout design rule checking (DRC, design Rule Check) relies mainly on manual adjustments by engineers to empirically trade-off area, timing, etc. The reliability of manual repair is high, but the efficiency is low, the repair quality is limited by personal experience, and the repair efficiency is difficult to ensure in a very large-scale layout. The automatic repair script can locally adjust DRC violations such as line width, package and the like in the layout through predefined rules. At present, the automatic repair modes mainly comprise the following two modes: One is for engineers to write automated scripts for digital IC back-end DRC repair, based primarily on predefined rules, that attempt to eliminate DRC errors using a series of fixed graphic operation templates. However, due to the limited statics and coverage of the rule base, the rule base is difficult to adapt to the illegal scene of the complex layout, and the repeated iterative repair or the problem that the rule base cannot be converged often occurs. And secondly, the DRC hotspot prediction tool predicts the probability of violating the DRC after the current layout is routed by using a convolutional neural network [1] [2] or a graph neural network [3], so that the probability of occurrence of DRC errors is reduced by optimizing the initial layout or routing, but the real DRC errors in the generated layout cannot be processed, and the GDS2/OASIS layout file cannot be directly operated. However, in either method, the limitation is that the rule base has limited statics and coverage, and is difficult to adapt to the illegal scene of the complex layout, and the repeated iterative repair or the problem that the rule base cannot be converged often occurs. And the research adopts a genetic algorithm, simulated annealing and other global optimization algorithms to search a feasible solution, but has the defects of high calculation complexity and low convergence rate. Reference is made to: [1]LIANG R J,XIANG H,PANDEY D,REDDY L,RAMJI S,NAM G J,HU J.Design rule violation prediction at sub-10-nm process nodes using customized convolutional networks[J].IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems,2022.https://ieeexplore.ieee.org/document/9608975. [2]LIN J G,CHEN Y G,YANG YW,HUNG W T,TSAI C H,FU D S,CHAO M C T.DRC violation prediction withpre-global-routing features through convolutional neural network[C]//Proceedings ofthe Great Lakes Symposium on VLSI 2023(GLSVLSI'23).2023: 313-319.https://doi.org/10.1145/3583781.3590216. [3]BAEK K,PARK H,KIM S,CHOI K,KIM T.Pin accessibility and routing congestion aware DRC hotspotprediction using graph neural network and U-Net[C]//2022IEEE/ACM International Conference On ComputerAided Design(ICCAD).2022.https://ieeexplore.ieee.org/document/10070024. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a reinforced learning-based GDS2/OASIS layout DRC violation repair method, which takes a GDS2/OASIS format layout as a reinforced learning environment foundation on the premise of not converting the GDS2/OASIS format layout into an image format, and establishes an operable and feedback layout repair environment in a real physical coordinate space by analyzing a target layer graph in the layout, thereby having stronger universality, precision and expandability and effectively improving the automation and intellectualization level of the back-end design. The invention relates to a reinforcement learning-based GDS2/OASIS layout DRC violation repair method, which comprises the following steps: Reading GDS2/OASIS layout files, and extracting a graph set of a target layer; calculating DRC violation number in a GDS2/OASIS layout through a DRC detection module to obtain a region to be repaired in the GDS2/OASIS layout; Converting a graph set of the region to be repaired in the GDS2/OASIS layout into a state representation of the layout; Constructing a repair strategy module for generating a graph adjustment action according to the state representation of the layout; and applying the graph adjustment action to a target graph in the GDS2/OASIS layout. Preferably, the repair policy module includes the steps of: Firstly, extracting a current layout state s t from a state representation of the layout; Step two, defining an action space A, extracting an action a t from the action space A according to the current layout state s t, and executing the action a t to obtain a new state s t+1; thirdly, calling the DRC detection module to obtain the DRC violation number after exe