ANN-Based SRAF Insertion for IC Layouts

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional methods for inserting sub-resolution assist features (SRAFs) into integrated circuit (IC) layouts are time-consuming and lack a predictable, cost-effective way to automatically select optimal insertion sites, especially for features sized less than 22 nanometers, leading to uncertainty in manufacturing and increased risk of defects.

Innovation Solution

The use of an artificial neural network (ANN) to generate usefulness maps for target features, allowing for the automatic selection and insertion of SRAFs into IC layouts, based on existing SRAF usefulness maps in a library, thereby improving printability and reducing manual inspection and mathematical model reliance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional manual inspection and mathematical models are used for SRAF insertion, then the process can be performed, but the runtime is time-consuming and productivity is low

Engineering Contradiction:

ImproveSRAF insertion speedVSAvoidruntime for SRAF insertion

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent replaces manual inspection and conventional mathematical models with an artificial neural network system. The neural network automatically learns patterns from training data and predicts optimal SRAF insertion sites, eliminating the need for manual analysis and reducing computation time significantly compared to traditional mathematical models.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs preliminary training of the neural network on a comprehensive dataset of IC layouts and their corresponding SRAF configurations before actual production use. This pre-training phase allows the model to make rapid predictions during deployment without requiring manual analysis or complex real-time calculations.

Inventive Principle:

Principle #10Preliminary action

2Manufacturing precision

If conventional mathematical models are used for SRAF insertion, then the process can be completed, but the quality of SRAF placement is insufficient and manufacturing precision is compromised

Engineering Contradiction:

ImproveSRAF placement accuracyVSAvoidpredictability of manufacturing outcome

Core Design Contradiction:

Manufacturing precisionVSReliability

Solution Approach 1:

The patent replaces conventional mathematical models with an artificial neural network that processes complex patterns from training data. The neural network captures non-linear relationships and subtle patterns in IC layouts that traditional mathematical models cannot adequately represent, leading to more accurate and reliable SRAF placement predictions.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system incorporates feedback mechanisms where the neural network continuously learns from actual manufacturing outcomes and adjusts its predictions. This feedback loop allows the model to improve its accuracy over time and adapt to variations in manufacturing processes, enhancing both placement precision and outcome predictability.

Inventive Principle:

Principle #23Feedback

3Ease of operation

If manual inspection techniques are used for SRAF insertion, then the process can be performed, but the ease of operation is reduced and device complexity increases

Engineering Contradiction:

Improveautomation level of SRAF insertionVSAvoidsystem complexity for SRAF insertion

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The neural network system performs self-service by automatically identifying optimal SRAF insertion sites without requiring manual inspection or intervention. The system processes IC layouts autonomously, making decisions about SRAF placement based on learned patterns from training data, thereby simplifying operation while managing complexity through automation.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The artificial neural network acts as an intermediary between the IC layout design and the SRAF insertion process. It translates complex design requirements into actionable insertion recommendations, mediating between the need for high precision and the desire for ease of operation by providing automated, data-driven decisions.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS10198550B2SRAF insertion with artificial neural network

Publication Date: 2019.02.05 TAIWAN SEMICONDUCTOR MANUFACTURING CO LTD

AI summary

Embodiments of the disclosure provide a method including: identifying a target feature in an integrated circuit (IC) layout not represented in a library, the library including a plurality of sub-resolution assist feature (SRAF) usefulness maps corresponding to a plurality of features and SRAFs in the IC layout; generating a usefulness map for the target feature with an artificial neural network (ANN), the generating being based on the target feature and the plurality of SRAF usefulness maps in the library; adding the target feature and the generated usefulness map to the library; selecting an SRAF insertion site for the target feature based on the generated usefulness map; and inserting an SRAF for the target feature into the IC layout at the selected SRAF insertion site.