Calibre intelligence

Calibre offers designers multiple model-based checking options to fit in any design flow: 

Prepacked models: trained with foundry data and deployed as a ready to run solution for the end user. 

• The Calibre LFD tool provides a virtual fab that predicts the “as manufactured” lithographic process variability impact on a design. It leverages machine learning to achieve 100X reduction in run times compared to first-principal models without sacrificing accuracy. 

Adaptable models: Pretrained by Calibre and fine-tuned with customer data to deliver flexible model-based checking capability for optimizing design manufacturability.   

• The Calibre SONR platform’s ML-Based process aware layout analytics tools enable full chip hotspot prediction and analysis, pattern reduction, and coverage checking in the design flow.  

DIY Custom models: Comprehensive feature extraction and machine learning tools are embedded in the Calibre infrastructure​ to train and deploy custom model-based applications within a Calibre verification flow. 

• Custom user-built models can be used for a wide range of applications.  
• One example application is enhancing pattern-based hotspot detection with a custom ML-model where the customer was able to achieve 21% improvement in accuracy compared to their traditional pattern matching flow.  

Calibre has embedded AI technology for intelligently utilizing hardware resources during a run.     

• The Calibre nmDRC core engine analyzes each layout and PDK to dynamically identify and select the best hierarchy optimization technique to use for optimal job distribution enabling peak scalability across hardware nodes.  
   
• The Calibre PERC reliability verification tool has ML-optimized resource distribution for PEX flows improving runtime by up to 400%. 

Expert guidance for the optimal Calibre setup adapted to the layout, process design kit (PDK), and user environment and objectives​. 

• The Calibre nmDRC Recon intelligent automation tool for verification runs early in the design cycle achieves an 82% run time improvement across 18 testcases.  
• The Calibre nmLVS Recon intelligent automation tool for early-stage circuit verification increases the number of LVS iterations users can complete in one day by up to 65X. The nmLVS Recon ERC functionality shows a 16X runtime improvement for ERC checks.  
• The Calibre nmLVS intelligent Hcell generation capability eliminates the frustrating process of manually optimizing Hcell selection and shows 10% improvement in runtime and a 30% improvement in memory.  

Accelerate the DRC debug process through intelligent results analysis tools that cluster errors to increase the probability of a designer identifying a repeatable fix or systematic design issue.   

• Technology-agnostic clustering that groups related errors across checks. 
• Context similarity clustering that groups related errors within checks.  

Stay tuned for Calibre’s new advanced visualization engine that will launch in 2025!  

• Fast and easy analysis of chip-level results across the die to quickly identify error hotspots. 
• Enables cluster-based guided debugging to simplify the debug process and accelerate DRC closure.  
• Numerous additional functionalities that assist in isolating, investigating, and identifying systematic design flaws.  

The Calibre DesignEnhancer layout optimization tool combines intelligent design optimization techniques with Calibre’s core verification engine for automated drc-clean design-stage IC optimizations.  

• Maximize via insertion with DE Via to achieve an average of 50% EMIR improvement.  
• Enhance the power grid design with DE Pge to achieve an average of 30% EMIR improvement. 
• Automatically place DCAP, ECO and filler cells to achieve a 15X run time improvement over traditional methods.  

Neural net assisted modeling for resist, mask and etch optimizes litho-patterning​ model accuracy while​ maintaining performance. 

• The optimized models are simpler, which helps reduce OPC runtime.  
• The model accuracy meets A18 and A14 requirements. 

Applying AI to OPC and RET solutions to improve turn-around time.  

• The AI in Calibre computational lithography products improve runtime by 2X to 4X while preserving the accuracy of full model-based OPC.  
• Monotonic machine learning speeds inverse lithography (ILT) by ~230X over a full ILT run and earns an F1 score of over 80% on N2 style logic layouts. 

ML-based checking for hotspot prediction/detection 

• Using machine learning in Calibre's litho-friendly design (LFD), users see a 100% run time improvement with 99.7% of the design for manufacturing (DFM) hotspot predictions within 0.5 nm accuracy. The Calibre LFD tool gives designers hotspot prediction and analysis independent of the fab's RET. 
• The Calibre SONR feature-vector driven machine-learning platform for hotspot prediction improves defect hit rate by 40X. 
• Defect classification and model coverage checks use machine learning to assure model accuracy required for A18 and A14 and prevent overfitting. 

Process driven, design content informed, yield debugging and process optimization.  

• Calibre SONR clustering and down sampling leverages AI for high-confidence target selection within the metrology budget and reliable hotspot prediction and detection. 
• Machine learning based automatic defect classification eliminates manual classification effort for reliable and repeatable results. 
• Calibre Fab Insights leverages AI for yield forecasting and debugging, DOE setup for solutions exploration and preventive maintenance prediction. Users see a reduction of metrology effort. 

Faster process development using intelligent pattern generation.  

Combining Calibre SONR and LSG Synthetic Layout generation technology offers:  

• Full coverage design rule exploration ​ 
• Test patterns for process development​ 
• Test pattern for PDK quality assurance​ 
• Exploration of context dependent standard cell behavior 

The best-in-class Calibre Cluster Manager offers predictability and efficiency in tape-out operations.  

• Automatically manages Calibre jobs with dynamic resource allocation to improve both compute cluster utilization and overall runtime for production tape-out operations. 
• Using AI, system anomalies are detected in real-time helps identify issues with your hardware cluster.