Tessolve AI Strategy

Tessolve’s AI Center of Excellence (CoE) follows a structured strategy to continuously train and improve AI models through iterative learning and validation. The process begins with a rich training dataset, supplemented by recorded user prompts to refine AI responses. The first trained version, Model v1, is developed and undergoes a validation phase using automated test data to assess accuracy and reliability. The validated model generates results, which are analyzed for performance improvements. Insights from this phase guide the infusion of Untrained Model v2 which is new to market but with better efficiencies. The new model undergoes further training and validation to enhance accuracy and adaptability. This even becomes better with recorded prompts and responses from the earlier model. This iterative cycle ensures continuous learning, optimizing AI models to handle real-world design verification and automation challenges. By leveraging automation and AI-driven validation, Tessolve enhances efficiency, minimizes errors, and accelerates verification workflows. This strategy establishes a robust AI framework, making AI-assisted verification more reliable and scalable.


Figure 1: Tessolve CoE AI Strategy

Enhancing Design Verification with AI-Driven Automation

Tessolve is redefining the landscape of semiconductor design verification by integrating artificial intelligence (AI) into its process, according to Bartley et al., (2024). Traditional verification methods often require extensive manual effort, which can be time-consuming and prone to human error. Tessolve’s AI-driven tools automate the generation of test benches, enabling designers to validate their RTL designs quickly and efficiently. By leveraging AI, the verification process is streamlined and made more reliable, reducing the chances of undetected errors in early design stages.


Figure 2: AI Assisted Design Verification Flow

The diagram represents the Design Verification Flow, showcasing how AI, Automation, and Machine Learning (ML) enhance verification efficiency. It begins with Requirements/Specifications, where AI assists in Feature Extraction to generate a Feature List. AI also aids in Coverage Planning, ensuring a well-structured Coverage Plan. Automation scripts handle Testbench (TB) and UVC generation, reducing manual effort and improving test quality. AI further automates Test Generation and Constraint Generation, ensuring comprehensive verification. The generated tests feed into Simulation, Debug, and Regression Suites, where ML optimizes the results. ML-driven analysis enables faster Coverage Closure, reducing verification cycles. AI engines ensure efficiency, accuracy, and adaptability across the verification process. The integration of AI and automation minimizes human intervention, accelerating time-to-market. This structured flow streamlines verification, ensuring robust chip validation.

Independent Verification for Robust Design Validation

Ensuring independent verification is a crucial step in design validation, as it eliminates biases and potential blind spots from designers, as highlighted by Bartley et al., (2024). Tessolve’s AI tools help generate test benches from RTL specifications in Verilog, System Verilog (SV), or Universal Verification Methodology (UVM), ensuring a structured and systematic approach to verification. Designers can integrate key verification components like assertions, functional coverage, and scoreboard functions, which verification engineers can choose to reuse or modify as needed. This approach fosters a collaborative verification process while distinguishing between design and verification roles.


Figure 3: Tool for AI generated Testbenches and Integrated Simulation with Open-Source Tool

Coverage Optimization with Machine Learning

One of the key challenges in design verification is achieving high coverage with minimal effort, as mentioned by Bartley et al., (2024). Tessolve employs machine learning (ML) techniques such as Gradient Descent and Genetic Algorithms to optimize verification goals. AI ensures that RTL designs achieve better functional coverage by analyzing test results and iteratively improving test scenarios. This intelligent approach reduces verification cycles and enhances overall test effectiveness, allowing engineers to focus on more complex validation tasks.


Figure 4: Coverage Improvement using ML

Seamless Integration and Accelerated Debugging

Tessolve’s AI-driven tools seamlessly integrate with existing tool chains, enabling smooth compilation and simulation workflows, as emphasized by Bartley et al., (2024). The built- in error feedback system allows real-time debugging, helping designers identify and fix issues faster. By automating smoke tests, designers can independently verify their designs without waiting for dedicated DV teams, reducing verification bottlenecks and accelerating the development cycle. Combining automation and intelligent debugging leads to significant time and resource savings for semiconductor companies.

The Future of AI in Semiconductor Verification

According to Bartley et al., (2024), Tessolve develops its AI strategy based on adaptability and continuous improvement principles. By incorporating replaceable AI models, centralized AI services, and customizable solutions, Tessolve ensures that its tools remain at the forefront of innovation. The company’s commitment to leveraging open- source technologies provides the industry with cost-effective and flexible verification solutions. Tessolve is revolutionizing how semiconductor companies ensure their RTL designs’ accuracy and reliability with its cutting-edge design verification approach as AI continues to evolve.

Note: To learn more about Tessolve’s AI-driven verification solutions, visit our website: www.tessolve.com.

References

Bartley, M., Soni, M., C Tessolve. (2024). AI Strategy for DV Flow & TB AI Tool. https://www.tessolve.com/

Author

  • Mike Bartley Tessolve

    Dr. Mike Bartley, Senior Vice President, Tessolve Mike is responsible for VLSI sales for Europe as well as supporting Design Verification (DV) sales globally. Mike has 30+ years DV experience and has built and managed state-of-the-art DV teams inside several companies (including STMicroelectronics, Infineon, Panasonic, and the start-up ClearSpeed) and also advised several companies on organisational DV strategies (ARM, NXP, and multiple start-ups). Mike successfully founded and grew a DV services company to 450+ engineers globally, delivering services and solutions to over 50+ clients in various technologies and industries. The company was acquired by Tessolve Semiconductors in 2020. Mike also manages the Centre of Excellence (CoE) activities at Tessolve. This includes building new products and services across all of the Tessolve business units (BU): VLSI, Silicon Test, PCB and embedded SW and products. The CoE is also developing AI capabilities and expertise across Tessolve. This is used to improve efficiencies across all the internal divisions (finance, HR, recruitment, procurement, program management, etc) as well as efficiencies and new services across the BU.

  • Marmik Soni Tessolve

    Marmik Soni, Sr. Design Lead, Tessolve Marmik Soni is a seasoned semiconductor industry professional with over 15 years of experience, currently serving as the Senior Design Lead at Tessolve's Center of Excellence (CoE). In this role, he leads the AI team in developing innovative AI and Machine Learning solutions across multiple business units. As an experienced Generative AI (GenAI) Strategist and AI Solution Architect, Marmik possesses a robust background in VLSI design. He is also a certified expert of ISO 26262 Automotive Functional Safety. Throughout his career, he has spearheaded chip design research and development, effectively managing large engineering teams. Under his leadership, Tessolve's CoE has developed over 25 AI-driven tools, positioning the company at the forefront of technological innovation. His expertise includes UVM testbench automation, assertion generation, and coverage closure, underscoring his commitment to advancing semiconductor design and verification through AI. Marmik has also been instrumental in securing research funding, mentoring scholars, and delivering expert talks at prestigious institutions. In recent years, he has delivered over 15 presentations on AI integration at various chip design conferences and summits. His work at the CoE has enhanced efficiencies, reduced cycle times, and fostered innovative AI solutions across business units and internal divisions, including finance, HR, recruitment, procurement, and program management.