After the Internet bubble burst in 2001, many people were full of doubts about the future development of the entire semiconductor industry.
In the round of market collapse at that time, many semiconductor companies began to integrate; the industry's attractive investment in wind capital was also greatly reduced; technology research and development in process development and other aspects have also stagnated and slowed down.
However, the semiconductor industry has seen a new turnaround now. In an interview with reporters such as Ji Wei.com, Mentor IC EDA Executive Vice President Joseph Sawicki said that the industry is re-filled with opportunities under the stimulation of new technologies such as artificial intelligence and machine learning.
A McKinsey report pointed out that artificial intelligence can be applied to many vertical areas, which allows semiconductor companies to capture 40 to 50% of the total value from these technology stacks. Joseph said that artificial intelligence will be a strong catalyst for another 10-year growth cycle in the semiconductor industry. But to make this trend truly realized, a lot of data is needed as a support.
“With enough data, you can be predictive, so you can train your machine very reliably and let the machine learn effectively.” Joseph further added that the amount of data needed and created for high-speed communication will increase over the next 12 years. It will usher in thousands of times of growth, and these data need to be analyzed, and then take action based on this analysis.
However, under the impact of the “data tsunami”, the development of artificial intelligence is also facing various contradictions. Joseph mentioned two conflicting goals in the development of artificial intelligence:
One goal is that many people want to continuously strengthen the capabilities of the data center to cope with such huge amounts of data. So companies like Alibaba and Amazon are developing AI-related engines that use this engine to train massive amounts of data.
On the other hand, the goal of some companies is to push more and more processing power to the edge of the cloud, thus releasing some pressure on the development of the data center.
Chip development in edge computing will greatly exceed the chip required by the data center. According to Tractica, from 2016 to 2021, the compound annual growth rate of edge-connected devices will be as high as 190%.
Joseph said that, closer, edge computing/processing will be the main engine for growth in the semiconductor industry. Since specific applications in many areas require optimized chip designs to achieve optimal chip performance, this will be an opportunity for EDA tool vendors like Mentor.
Joseph emphasizes that in edge computing AI, chip design is often defined by specific architecture development requirements. So the current AI development platform is completely different from the previous development environment.
In this regard, Joseph introduced Mentor's chip design tools specifically for the AI field:
lHLS (high-level synthesis): Take NVIDIA as an example. By using this tool, you can increase productivity by nearly two times and verification costs by 80%.
lHierarchicl test: Helps customers further increase productivity and reduce costs. Taking Graphcor's customer as an example, by using this tool, DFT productivity has been increased by 4 times, the speed of test transfer has been greatly improved, and the design time period has been shortened to 3 days based on actual data.
lOPC technology: used in semiconductor manufacturing, it takes 4,000 CPUs to run one day on a 7nm basis to produce one Mask, but if you use machine learning algorithms, you can reduce the running time by 3-4 times.
lLFD (lithographically friendly) technology: significantly reduces the yield limit factor and reduces the run time of 10 times production. Not only can identify defects in the production process, but also predict defects.
lDeposition tool: solves the problem of product or component failure and improves the quality and efficiency of production.
In addition, Mentor provides a characterization technology platform for the automotive industry, providing a detailed analysis of the overall reliability and safety, combined with AI to reduce the runtime of characterization by a factor of 100. The PAVE 360 Autopilot Simulator also continuously simulates real-world conditions under the virtual machine, further reducing verification time.
Whether the future smart chips are dedicated or flexible, the industry has different voices. But Joseph told the micronet reporter that EDA is a neutral tool. In the future, Mentor will provide a large environment where customers can use the tools to model and develop their software in specific environments. This is the most important value that Mentor offers as an EDA company.