Solido Launches Machine Learning (ML) Characterization Suite
Growth in quickly advancing semiconductor segments, including high performance computing, automotive, mobile, and Internet of Things (IoT), is driving chip complexity with the need to move to more advanced designs. Standard cell, memory, and I/O characterization is a resource-intensive runtime bottleneck, often with challenges to meet production accuracy requirements.
Solidos ML Characterization Suite significantly reduces the time and resources required for library characterization, while delivering production-accurate library models. ML Characterization Suites machine learning algorithms efficiently and accurately model the characterization space. Real-time cross-validation techniques are then applied to determine model error and fit. Solidos approach is tuned to work across all Liberty data types, including timing, power, noise, waveform, and statistical data.
The following tools are included in ML Characterization Suite:
- Predictor instantly and accurately generates new Liberty models at new conditions. It does this by modeling the full Liberty space using sparse data from existing Liberty models at other conditions. This reduces up-front characterization time as well as turnaround to generate Liberty files. Predictor works with all Liberty formats and reduces library characterization time by 30% to 70%.
- Statistical Characterizer quickly delivers true 3-sigma statistical timing data (LVF/AOCV/POCV) values with Monte Carlo and SPICE accuracy, including non-Gaussian distributions. It adaptively selects simulations to meet accuracy requirements and to minimize runtime for all cells, corners, arcs, and slew-load combinations.