Helping chip design with AI, Mentor has a coup
Published time: 2019-09-11
“The semiconductor industry has experienced many setbacks before. Since the Internet bubble burst in 2001, many people have been confused about the future development of this industry. After the market crash, many companies began to integrate. Our industry is in process development, cycle. R&D in other areas has also been stagnant and slowed down. But fortunately, we are now seeing a turn for the industry.” Joseph Sawicki, Executive Vice President of Mentor IC EDA, was on the Mentor Forum on August 28, 2019. Said, enough to see how many twists and turns in the semiconductor industry.
Fortunately, the turnaround appeared.
Joseph Sawicki continues, this turn is brought about by artificial intelligence and machine learning related applications, and with the stimulation of these technologies, we see that the industry is once again full of opportunities. According to a McKinsey report, artificial intelligence is opening up opportunities in the semiconductor industry for decades, because artificial intelligence can be applied to many vertical markets. For example, in the mobile field, the semiconductor industry can capture the 20 created by the industry. %the value of. At the same time, the research report from PWC also pointed out that artificial intelligence will be a very strong development catalyst for our other 10-year growth cycle. The report also pointed out that among this round of opportunities, companies that can make the most of the possibilities and opportunities brought about by artificial intelligence will benefit the most.
As a leader in electronic design automation, Mentor is committed to providing complete software and hardware design solutions that enable companies to develop better, more cost-effective electronic products. When the shareholder of AI blew, Mentor took the lead in smelling the opportunity and using AI to improve the effectiveness and efficiency of the tool.
According to Joseph Sawicki, the most fundamental thing about AI is relying on big data. If you have enough data, you can be predictive, so that you can train the machine very reliably and let the machine learn effectively. For the large amount of data collected, analyze it and take action.
What does AI bring to the semiconductor industry?
Under the impact of such huge data, there are various contradictions in the development of AI. On the one hand, the huge amount of data makes many people want to continuously strengthen the capabilities of the data center, so many companies are now developing artificial intelligence-related engines, using this engine to train and train massive amounts of data. On the other hand, some companies have set themselves the goal of pushing more and more processing power to the edge of the cloud, which can release some pressure on the development of the data center.
The two directions are contradictory but reasonable. In Joseph Sawicki's thinking, chips in edge computing seem to be growing faster. He cited an example of a small wind farm that can be used for preventive maintenance of wind turbines or wind turbines. For the offshore wind mill, the operating environment it faces is very demanding, and the cost involved in maintenance is very high, so the advantages brought by edge computing are very obvious. By performing some preventive analysis on-site at the wind farm and then pushing these results to the data center, the cost of maintenance can be saved by 64%. Mentor captures the opportunity in the customer's needs, and by developing application-specific chips, it is easier to capture the potential value of the artificial intelligence market.
Speaking of chip design, Joseph Sawicki mentioned that there are many areas where specific applications require optimized chip design. Therefore, design plays a crucial role. What can AI bring? Generally speaking, when designing SoC, SoC design is driven by specific specifications, such as the number of cores and processing power, etc., and now in the edge computing AI, The design of the chip is often defined by specific architectural development requirements. So the current AI development platform is completely different from the previous development environment. For example, HLS (High-Level Synthesis) is a 20-year-old technology. The methods used in the past have been very limited, and most of them are used for the implementation and execution of Data Path. But this technology has become crucial in the AI era, it is related to the implementation and implementation of the entire AI platform, from top to bottom. It is necessary to set the performance, size, memory configuration, etc. of the entire system in the entire algorithm development environment. According to an article published by industry leader Nvidia, using this tool can increase productivity by nearly two-fold and verification costs by 80%.
At the same time, the testing and verification involved in AI-related SoCs is very different. In the traditional sense of doing SoC, you need to verify and compare the specifications it needs to achieve and its functions. What is the interconnectivity? Of course, this is still important in the AI era, but more important is the verification of performance, the verification of power, and the verification of the implementation of the entire architecture. Only after these have reached the ideal state can the designed chip be optimized for the vertical market it faces. To achieve this, Mentor simulates in a virtualized environment, simulates the real environment with a chip, and verifies it in a real-world environment. This includes verifying MLPerf and so on to verify specific results. In order to ensure that the chip can finally meet the needs of customers, in order to optimize the algorithm for executing AI, Mentor needs to greatly improve the performance of the chip.
What does AI bring to the EDA industry?
AI and machine learning will bring changes to the EDA industry while bringing changes to the semiconductor industry. Joseph Sawicki further introduced that Calibre, which is the industry leader in physical verification, can be used for DFM (designed for manufacturing) applications, and designers can use this software to determine that their designs can ultimately be Execution, being produced. At the same time, wafer fabs can use this software to help them create chips with optimized performance. Shape recognition is verified by these software and related artificial intelligence features. For example, there are many different shapes on the chip. Through such a function recognition and shape recognition, it is possible to analyze some trends based on big data, and then based on this insight to better deal with data that has not been trained.
Let's talk about OPC technology, which can be applied to semiconductor production, and customers can control the output well in its design. It is very complicated. On the basis of 7nm, it takes 4,000 CPUs to run one day to produce one Mask. If you use machine learning algorithms, the whole running time can be reduced by 3-4 times, which can help customers save a lot of cost. .
There is also a "lithographically friendly" design technique. With this technology, the limiting factor of yield can be greatly reduced. Machine learning can greatly reduce the running time of real production. This is an image-based technology that not only identifies defects in the production process, but also predicts them. This technology can greatly increase production yield.
Let me talk about a case for the diagnosis of test results. Joseph Sawicki believes that a product or component failure or failure is an opportunity for Mentor. Through the machine learning, the entire design, production records and other aspects can be analyzed in detail and the results can be obtained. If the results of the operation show a 5% loss in the entire production process, the Deposition (deposition) tool can be used to solve this problem throughout the production environment, improving the quality and efficiency of the entire production.
Characterization techniques are also very important, especially for the automotive industry, which is highly demanding for reliability and safety. Traditionally, the Monte Carlo method is used to analyze the entire reliability and security, but it involves too many simulations from the point of view of calculation, so it is very impractical. Based on Mentor's second and third generation technology platforms, combined with AI, the runtime of characterization can be reduced by a factor of 100. A critical analysis of 7 Sigma is very valuable for the automotive industry.
Mentor seized the opportunity of AI and developed a unique path of his own. Joseph Sawicki concluded that, like Linux, everyone is using Linux now, and in a few years, everyone will use AI. So the most important point is which application you are using for the AI, so by the end everyone will use the same AI, and the tools developed based on AI will be different. Mentor is such a company that provides methods and tools to help customers actually implement the chip.
- 1.Common fault analysis and maintenance methods for power amplifiers
- 2.LED drive power classification and characteristics
- 3.Application Analysis of Light Emitting Diodes
- 4.Wind speed sensor installation requirements and methods
- 5.Apple mobile phone chips go down the altar?
- 6.Gas sensor application field and development prospects
- 7.Introduction to the types and characteristics of PIC microcontrollers
- 8.The performance of the water level sensor is broken and the method of judging the quality
- 9.How to determine reasonable radiation intensity and collector current for photoelectric sensors
- 10.Principle and dynamics of wireless pressure sensors
- 11.Helping chip design with AI, Mentor has a coup
- 12.How does the human proximity sensor work and what it can do?
- 13.Introduction to common measures for suppressing interference sources in PCB circuits
- 14.Characteristics of series and parallel circuits
- 15.Monitoring method for partial discharge of power transformer