Artificial intelligence as a tool to create processors: Samsung leads by example
Artificial Intelligence technology is becoming more and more popular. It’s not just a tribute to fashion but actually working projects that help realize various ideas. Now AI technologies are also penetrating into electronics manufacturing and development, including the most valuable of all: processors.
Creating modern chips is extremely difficult, a very long, multi-step process. Companies such as Intel, Samsung, Apple, AMD, Chinese manufacturers manage (at least for now). But if the process can be optimized and accelerated, why not? And this is where artificial intelligence comes in. Samsung has even developed the first commercial samples of chips created with the help of AI.
Who developed the technology and when?
In truth, the credit doesn’t go to Samsung – that company only took advantage of a third-party solution. But a pioneer in the use of AI for chip design was the company Synopsys.
It is known primarily for its main line of business: CAD for electronics design. Its most famous product is Design Compiler, a synthesis tool for logic circuits. But there are other products, which are also used for the development of specialized integrated circuits.
Thus, the company offers partners products for logic synthesis, behavioral synthesis, element tracing, static time analysis, formal verification, HDL simulation (SystemC, SystemVerilog/Verilog, VHDL), and circuit simulation. Simulators include development and debugging environments.
Well now, it seems that Synopsys developers have outdone themselves and learned how to use artificial intelligence to create circuits.
How it works
Synopsys does not disclose all the details of its solution, but it is known that DSO.ai, Design Space Optimization AI helps to determine the optimal placement of silicon elements on the chip to reduce the overall area and reduce energy consumption to increase productivity.
During analysis, DSO.ai evaluates billions of different alternatives and creates a final variant that compares favorably with the rest. The developer of the technology claims that the variant proposed by the system compares favorably even with those proposed by human experts.
When designing a processor, there are about 1090000 possible ways to place components. This is comparable to the number of moves in the game Go, which artificial intelligence has already mastered. Of course, processor design is not a game, but still, the system needs to choose one correct option among a huge number of not-so-good ones. And it works – the frequency of the processor the system designs is 18% higher on average with a 21% reduction in power consumption. Plus the development time is reduced from six months to just one.
The main beneficiary, so to speak, was the South Korean company Samsung, which was given the opportunity to work with this technology. The South Koreans did not fail to take advantage of the offer, developing the first sample of a commercial processor in a short time.
This is not at all a small test project, which has not yet left the R&D stage – Samsung has already confirmed that it is using new software from its partner to create real chips. The novelty helps increase the efficiency of the development of Exynos chipsets, Samsung’s proprietary products. True, to confirm – confirmed, but the details at the same time reported a minimum.
In particular, it is not known how many processors are produced, developed with the help of new technologies, whether the production is mass production and, if so, in what systems Samsung all this is used. True, we can try to guess. The other day, the corporation unveiled its newest foldable smartphone, the Galaxy Z Fold3 – just the latest Exynos is installed inside. And who knows, maybe this week Samsung will announce that the new process has been developed with the help of artificial intelligence.
Why exactly AI?
According to experts, including Mike Demler, a senior analyst at the Linley Group, AI is excellent at solving the problem of efficiently placing billions of transistors on a chip. Demler generally believes that artificial intelligence is the future of semiconductor manufacturing and development.
Nowadays AI has to be trained on the basis of large amounts of data. And this is not always possible in such a specific industry as the production of chips. If it is possible, it requires time and other resources, which are always in short supply in the semiconductor components industry. Accordingly, when working with AI, there is a question of money – even the simplest systems based on chips created by AI become very expensive.
In the near future, according to Demler, there will be no need to worry about training, because models for machine learning will become much more advanced than they are now. Should engineers start worrying? It is unlikely, the same expert says that it will not be possible to completely replace people with electronics.
Not just Samsung
As far as we can understand, the South Korean corporation was the first of many – and these many will not slow down soon. The fact is that the tool from Synopsys, which is called DSO.ai, will soon be provided to other major market players. And to be more exact – to a great many of them, because Synopsys works with a huge number of partners.
The tool has all the capabilities to speed up semiconductor chip development by adding and modifying various components, design elements, etc. Other companies are developing similar tools.
AI as a new trend in chip development
Synopsys was the first company to offer its development to partners, and they jumped on it. But there are other companies that are seriously considering using AI in chip development. These are, in particular, Nvidia and IBM. By the way, other companies, including Cadence, a direct and quite dangerous competitor to Synopsys, are now trying to develop AI technology for chip development.
Google announced that its new flagship has an AI-designed chip. In particular, AI helped create some important components of the Tensor chip. In addition, the company’s next smartphone, the Pixel 6, received a custom chip from Samsung. Unfortunately, a Google spokesperson did not comment on this chip – but there is a non-zero probability that this processor is just a new AI development.
A representative of the Linley Group analytical agency believes, not without reason, that the production and development of electronics using AI is the future. According to him, AI technology in the semiconductor industry will soon become the de facto standard.
In general, modern microprocessors are extremely complex, their development requires people who have been engaged in the industry for many years. It takes years, if not decades, to train high-level experts. And if such people are not found, and there is no reliable AI technology – the crisis of the semiconductor industry, which everyone is talking about now, cannot be prevented.
As for the AI technology itself, which was mentioned above, the companies that work with it use a machine learning method called reinforcement learning. Scientists say that this method works perfectly and they are not going to change anything for the time being. Well, the method itself will gradually be offered to dozens of companies – perhaps they will be happy to grab the idea.