![© 2019, AITS IEEE CS Chapter ai-techsystems.com
AI-DA: Opportunities
❑AI and EDA are both growing
❑ AI/ML to add $20 trillion in global economy by 2035. [Reference: PwC report]
❑ Design Automation Market to hit $14 billion by 2024. [Reference: GMI report]](https://image.slidesharecdn.com/mida-ieee-cs-milpitas-190924140147/85/Artificial-Intelligence-in-Design-Automation-5-320.jpg)
![© 2019, AITS IEEE CS Chapter ai-techsystems.com
AI-DA: Opportunities
❑AI/ML research rate is phenomenal
❑ Papers: well over 100k per year. [Reference: ScienceDirect APIs]
❑ Funding: AI/ML research is over 1% of total global research.](https://image.slidesharecdn.com/mida-ieee-cs-milpitas-190924140147/85/Artificial-Intelligence-in-Design-Automation-6-320.jpg)
Artificial Intelligence in Design Automation
- 1. IEEE CS Chapter, Milpitas Rohit Sharma
- 2. © 2019, AITS IEEE CS Chapter ai-techsystems.com About me Author, Entrepreneur and Technologist with 20 years in EDA/CAD industry Worked for TI, Magma, Apache, Cadence, Paripath and AITS. Passionate about software and technology. Busy creating deep learning framework and compiler linkedin.com/in/srohit0 medium.com/@srohit0 qr.ae/TWGSt9 github.com/srohit0 twitter.com/srohit
- 3. © 2019, AITS IEEE CS Chapter ai-techsystems.com Why is AI important ❔⁉❓ ❑ AI is the new electricity, new oil. ❑ AI is 4th industrial revolution ❑ Railroad ❑ Assembly Lines ❑ Digital ❑ Artificial Intelligence ❑ AI is becoming central tenet for the disruptive changes
- 4. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI Opportunities Accenture 2018: AI is the Future of Growth, reference: bit.ly/ai-growth A comparison of baseline annual gross value added growth (%) in 2035 to a scenario where AI has been absorbed.
- 5. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Opportunities ❑AI and EDA are both growing ❑ AI/ML to add $20 trillion in global economy by 2035. [Reference: PwC report] ❑ Design Automation Market to hit $14 billion by 2024. [Reference: GMI report]
- 6. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Opportunities ❑AI/ML research rate is phenomenal ❑ Papers: well over 100k per year. [Reference: ScienceDirect APIs] ❑ Funding: AI/ML research is over 1% of total global research.
- 7. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Opportunities ❑ The change will likely come in the form of product features, not new products or flows. ❑ AI features will bring better usability, accuracy and performance. ❑ AI won’t bring a disruptive change to EDA unlike other industries. Yes, EDA is ready to make use of AI research
- 8. © 2019, AITS IEEE CS Chapter ai-techsystems.com AITS Platform Mathematics Algebra Calculus Statistics Probability CAD Infrastructure Python DNNC Numpy Pyplot TCL CLI C++ EDA Algorithms Jupyter Google Colab … Proprietary Guna PASER Community apps Third party apps
- 9. © 2019, AITS IEEE CS Chapter ai-techsystems.com VLSI Cell Classification Train a machine learning model to classify the type of IP with a fully extracted netlist. Char Case Study: VCC Classification IP Classifier Graph Matrix Circuit graph
- 10. © 2019, AITS IEEE CS Chapter ai-techsystems.com ❑ VCC is a common design automation problem used for detecting patterns in ASIC, AMS and Custom flows. ❑ It is a pattern detection algorithm in its generalized form and applicable to many other problems including DRC, Extraction, Circuit Simulation and others. VCC Applications
- 11. © 2019, AITS IEEE CS Chapter ai-techsystems.com VCC Dimensionality Press F5 to start animation
- 12. © 2019, AITS IEEE CS Chapter ai-techsystems.com GUNA: AI enabled Char Platform ❏ Guna is built on AITS’s AI platform to offer machine learning in the cloud with VCC and other AI features. ❏ VCC repository dramatically improves ease of on pre- characterized cells. ❏ Helps with setup for re-characterization, multiple corners new process nodes. ❏ VCC feature adds to QA confidence of models.
- 13. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Challenges Clear Value Proposition Use Model Data Engineering High Dimensionality AI Technology Selection Integration into Legacy Systems Acceptance of Probabilistic Results Design Automation
- 14. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Applications ❑ Capacitance estimation ❑ Wire load Models estimation ❑ RC-tree estimation ❑ Early timing analysis ❑ RTL power analysis ❑ Cell and IP classification ❑ Verification Coverage ❑ Fast extraction ❑ Speed up circuit simulation ❑ Design Segmentation
- 15. © 2019, AITS IEEE CS Chapter ai-techsystems.com ❑ While EDA is busy catching up with ML/DL technologies, design community is busy designing AI chips. ❑ As of early 2019, AI chip market is $7 billion and expected to grow to over $66-$100 billion by 2025. AI Designs Sources: 1. alliedmarketresearch.com/artificial-intelligence-chip-market 2. tractica.com/newsroom/press-releases/deep-learning-chipset- market-to-reach-66-3-billion-by-2025
- 16. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI Chip Landscape (Aug 2019)source:github.com/basicmi/AI-Chip
- 17. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI smart appliances Smart Home Smart Industries Smart Cities Smart Infrastructure
- 18. © 2019, AITS IEEE CS Chapter ai-techsystems.com ❑ New IPs/designs will enable new wave of appliances with smart edge devices and controllers. ❑ Every new design will foster more jobs and opportunities. ❑ Compiler is one pervasive software needed to enable DL/ML models on these chip. ❑ Compiler skill is in high demand AI : Opportunities
- 19. © 2019, AITS IEEE CS Chapter ai-techsystems.com bit.ly/dnnCompiler It is an open source project to enable embedded systems providers a unified approach to compile models directly into the edge devices.
- 20. © 2019, AITS IEEE CS Chapter ai-techsystems.com ❑ World is focused on business with HPC devices like GPU/TPU, edge devices like VPU. ❑ Over 30 billion microcontroller-powered devices with low compute produced annually are being deprived from AI- revolution. ❑ DNNC plans to put ML/DL models into ignored set of devices. Why another compiler?
- 21. © 2019, AITS IEEE CS Chapter ai-techsystems.com Thank you That’s all folks !
- 22. © 2019, AITS IEEE CS Chapter ai-techsystems.com ❑ The VCC dataset consists of over 12,000 cells from multiple standard cell libraries with a variety of representative cells. ❑ RC data was stripped from fully extracted cell netlist and resulting netlist was represented as numbers. ❑ Cells were given 10 different labels. VCC Dataset
- 23. © 2019, AITS IEEE CS Chapter ai-techsystems.com AI-DA: Solution Flow Data Acquisition Data Analysis Model Hypothesis Training Validation Deployment
- 24. © 2019, AITS IEEE CS Chapter ai-techsystems.com Case Study: Ensemble K-nearest neighbors Bayesian networks Decision trees Boosted trees Logistic Regression Support vector machines Neural networks Random forests
- 25. © 2019, AITS IEEE CS Chapter ai-techsystems.com VCC Design Results … … … … … … … …