Expanding Silicon Lifecycle Management To Real-Time System Performance Optimization

Semiconductor development is currently in one of its periodic crises, with many factors combining to require dramatically new technologies and methodologies. Chips continue to grow ever larger and more complex, with 3D IC devices adding another layer of challenges. Huge data centers, autonomous vehicles, and algorithms using artificial intelligence (AI) and machine learning (ML) drive a relentless demand for maximum performance. Silicon cost and longevity are important considerations for the Internet of Things (IoT), where devices may remain in place for many years. Many applications also have high requirements for reliability, functional safety, and security, adding to the difficulty of the development process.

The upshot is that the old ways of designing, manufacturing, and deploying semiconductors no longer suffice to deliver what the end markets need. Incremental improvements are not enough to move forward; a new path is required. Silicon lifecycle management (SLM) has emerged as a holistic approach to address these challenges. SLM has two primary aspects:

• Deploying sensors, monitors and structures embedded within silicon designs to gain insight into how devices are made and how they perform in the field
• Gathering and analyzing data at every opportunity throughout the lifecycle of silicon devices to enable in-design, in-production and in-field device optimization and improvement

Synopsys has made a major commitment to SLM over the last few years with our SiliconMAX Silicon Lifecycle Management platform. Last week we announced the next step in our ongoing SLM investments with the acquisition of Concertio Inc., a provider of AI-powered performance optimization software. In this blog post we’d like to explain how the pieces fit together seamlessly to evolve and enhance our SLM solution.

Let’s start by looking at the big picture. SLM is a broad process that encompasses monitoring, analysis, and optimization of semiconductor devices as they are designed, manufactured, tested, and deployed in end user systems. The results of the analysis and optimization benefit semiconductor and system design, bring-up, manufacturing, and production test teams as well as the end users of the systems containing the chips. When fully deployed, SLM improves chip design, manufacturing yield, system performance, reliability, safety, security and more. The full solution has many elements and requires numerous tools and technologies to deliver the full range of benefits to all stakeholders. Delivery of this full solution is an evolutionary process, with new capabilities being added all the time.

Prior to this acquisition, our SiliconMAX SLM Platform has been focused on chip design, ramping up production and manufacturing test. The Platform includes several targeted analytics engines that operate on available chip data to enable optimizations at these early semiconductor lifecycle stages. We provide unique capabilities, with a vision that extends to optimization during bring-up and in-field operation. The current components of this solution include:

• Embedded agents, sensors and PVT monitoring IP
  • Including DesignWare PVT Monitors and Sensors
  • Including traditional Synopsys TestMAX DFT/BIST test IP
  • Real-time embedded sensing and thermal monitoring
• Fusion Design Platform
  • Direct links with the Fusion Compiler RTL-to-GDSII for optimal placement of the monitors and sensors while maintaining optimal design metrics
  • Silicon-aware design flow improvement
• Yield Explorer design sensitivity analysis engine
  • Manufacturing and test operational efficiencies for faster production ramp
  • Reduction in design-related yield limiters
• SiliconDash semiconductor manufacturing analytics engine
  • High volume analytics screening and production control to optimize manufacturing
• SiliconMAX High-Speed Access IP and TestMAX ALE Adaptive Learning Engine
  • Adaptive high-bandwidth testing
  • In-field access to monitor and sensor data

As in any AI/ML application, more data and more analytics foster deeper and better learning. The SiliconMAX SLM Platform performs big data analytics during design, product ramp and manufacturing test. Information from individual die and wafers is uploaded to a unified database over time. When many chips are being manufactured and tested, consolidating and analyzing the data enables ML to spot common issues, identify common solutions, share effective optimizations and identify trends over time. For example, harvesting parametric data across high volumes of production chips enables highly representative timing simulation models and the ML selection of Monte Carlo simulations. The Yield Explorer engine interprets the voluminous sensor data and passes only meaningful information to the analytics.

Our acquisition of Concertio adds capabilities to the SiliconMAX SLM Platform that extend its capabilities into bring-up and in-field operation. Concertio has developed unique AI-powered dynamic performance optimization software that improves system-level speed and resilience. An autonomous software agent installed on the target system continuously monitors the interactions between operating applications and the underlying system environment. The agent’s AI capability learns about the behavior of the applications through reinforcement learning techniques. Using this information, the agent’s optimization engine can adapt and reconfigure the system dynamically, resulting in a real-time, self-tuning system always optimized for its current usage. Current users have found that the agent can deliver a 5-15% improvement in overall system performance. Adding this form of optimization is a major step toward our vision for a complete end-to-end SLM solution, from chip design to end-user deployment.

There are numerous opportunities to link the elements of the expanded Platform in new and powerful ways. For example, linking the software agent to the hardware monitors and sensors will enable additional capabilities such as power optimization, early detection of aging effects and preventive maintenance before silicon failures can occur. Although the edge analytics in the agents are very powerful, over time many individual chip agents will be connected via the cloud to enable big data analytics and more proactive responses. For example, if a specific aging effect or other form of silicon degradation is being reported from many end systems, this is likely to lead to optimizations in the design, manufacturing, or test steps to improve chip robustness and reliability.

SLM is a powerful approach that makes dramatic advancements over traditional ways to design, build, test, bring up, and deploy semiconductors. Synopsys has a broad vision with many of the key components for SLM already available and proven on many projects. To learn more, a white paper is available. The addition of the novel technology from Concertio extends our current solution to encompass system-level performance optimization and sets the stage for many additional capabilities in the SiliconMAX SLM Platform going forward.