How to Accelerate Semiconductor Design Validation without Sacrificing Reliability
Semiconductor designers face pressures in many dimensions including integration, customization, and reliability. Balancing pressures with accelerated time-to-market goals is tough. Advanced analytics software enables designers to ensure reliability without time-to-market delays.
Pressures on Semiconductor Designers
Integrated circuits now have more transistors and functions than ever before. Carefully vetting each component and system requires more testing and analysis. Time spent verifying and validating can take up to 50% of the product design cycle. Automotive, IoT, and mobile device markets are demanding customized chips for each application. Custom work can require up to 12-18 months of debugging. Even though verifying, validating, and debugging take a lot of time, they are important to ensure reliability. Consumers are demanding reliability the first time otherwise they will go somewhere else.
Designers spend the bulk of their analysis in 2 phases: pre-silicon verification and post-silicon validation. In the pre-silicon phase, many test cases are run in simulation and lots of data is generated. Due to data complexities, it is difficult to explore all corner cases. In the post-silicon phase, tests are run on prototypes but it’s difficult to debug inside the design. Both phases have challenges, but designers agree that spending more time debugging in pre-silicon minimizes design risks later on. But how do designers handle all that data without slowing down the project?
Pre-Silicon Verification Engineers Overwhelmed by Data
Debugging pre-silicon produces a lot of data. There are tests and benchmarks to be performed at all stages. Today’s complex chips require physical simulation of pre-silicon logic usually using field-programmable gate array (FPGA) prototyping and system on a chip (SoC) emulation. Much time is spent waiting for simulations to run and data to transfer from instruments to servers. Large waveforms from oscilloscopes and analyzers can produce up to 20TBs of data. Large datasets can take up to 100 days to run and transfer. Engineers sometimes end up looking for anomalies in the data themselves. Manual analysis can easily consume up to 40 engineering hours.
Take for example an analog integrated circuit. In pre-silicon, validation engineers are looking to debug for voltage and current spikes, input/output glitches, and time shifts. It is time-consuming to emulate 64 channels while simultaneously looking for functionality and design. Each waveform per channel captured using an oscilloscope is a very large file. It can be in the area of 600 MB to TB of data for each. It’s just too hard to look through all the data to find meaningful patterns by hand!
Advanced Analytics Software Enables Deeper Insights
Analytics tools can replace guesswork and human intuition with fact-based knowledge, pattern recognition, and structured learning. Machine learning and artificial intelligence provide greater insight into diagnosing and preventing complex chip failures than humans. Advanced data analytics software can find anomalies in pre-silicon data.
Take for example waveform data taken from a Bluetooth low-energy device over a period of 17 hours. There are about 7 million waveform segments to inspect for anomalies. Let’s say that an abnormal current spike occurred only 17 times out of 7 million waveforms. The current spikes are impossible to notice without advanced analytics software. While uncommon, such abnormal current spikes can damage the electrodes of the battery. Current spikes can cause device malfunction due to fluctuations of the power supply voltage or noise interference.
Semiconductor design validation engineers using advanced analytics software can quickly identify and analyze the root cause of anomalies. Engineers can take corrective action before committing their design to silicon. They can create more reliable devices in less time. The improved reliability is especially important for devices used in mission-critical areas such as automotive and healthcare where human lives may be at risk. Advanced data analytics also brings competitive advantages because it allows engineers to spend more time innovating. It’s exciting to see semiconductor companies looking to advanced analytics software to accelerate their design validation.