Artificial intelligence (AI) is rapidly becoming the right-hand man for chip engineers in the extremely complex task of semiconductor design. According to Deloitte forecasts, the world's leading semiconductor companies will spend $300 million on in-house and third-party AI tools for designing chips in 2023, and will grow at a rate of 20 percent per year over the next four years to exceed $500 million by 2026. AI design tools enable chipmakers to push the boundaries of Moore's Law, saving time and money, alleviating talent shortages and leading old chip design into a new era.

Deloitte & Touche recently released the Deloitte TMT Predictions 2023, which explores the critical role that technology will play in a highly connected world. Among them, AI-designed future chips will be one of the trends to pay special attention to in 2023, while environmental sustainability will remain a key focus, and the technology industry will leverage its industrial strengths to achieve its net zero goal faster than other industries.

Semiconductor companies are turning to the use of AI and high-power materials to design future chips

Semiconductor companies are using AI technology to design chips with the goal of achieving faster, cheaper and more efficient. Deloitte predicts that in 2023, the world's leading semiconductor companies may invest $300 million in in-house and collaborative AI tools for chip design; and over the next four years, the $300 million invested will likely grow at an annual rate of 20% to exceed $500 million by 2026. The chips designed by AI software tools may be worth billions of dollars, far exceeding the cost invested by vendors, bringing significant excess return on investment.

Against the backdrop of a global semiconductor market expected to be worth $660 billion in 2023, the economic scale impact of AI will far exceed the cost of its design tools and enable chipmakers to push the boundaries of Moore's Law, saving time and money. At the same time, the use of AI tools can improve supply chain security and help alleviate the next chip shortage problem.

And high-powered materials made of chips will be suitable for electric vehicle (EV) batteries, solar panels, advanced military applications, etc. Deloitte predicts that total sales of high-powered chips will reach $3.3 billion in 2023, an increase of nearly 40% over 2022, and accelerate growth to nearly 60% in 2024, generating more than $5 billion in revenue.

Time is money: advanced AI tools to accelerate chip design

IC design has long exceeded the extent of manual design can handle, suppliers use electronic design automation (Electronic Design Automation, or EDA) to design and produce chips for decades, until 2022, the EDA industry will exceed $10 billion in value, and grow at a rate of about 8% per year. EDA tools typically use rule-based systems and physical simulations to help human engineers design and verify chips, and some even incorporate basic artificial intelligence. In the past year, Synopsys, the largest EDA company, has begun selling advanced AI tools, and chipmakers and technology companies have been developing their own AI design tools. Advanced AI tools are not just experimental, but are actually used in multi-billion dollar chip designs. While AI tools won't replace human designers, their complementary advantages in speed and cost-effectiveness give chipmakers a more powerful design capability.

AI is currently helping chip design and manufacturing in three main areas.

Making newer and better chips: chips below 10 nanometers are the fastest growing segment of the chip market, widely used in smartphones, computers and data centers, and by far the most profitable. Because the smaller the chip technology node, the higher the integration density of the chip, chips below 10 nanometers cost more than $500 million per new design, and they also have the highest manufacturing costs. Advanced AI tools can design chips faster than traditional methods, thereby reducing manufacturing costs.

Improving old chips: In 2022 all chips sold, two-thirds still use 65 nanometers or greater technology nodes, foundries must continue to introduce new process technologies, the old chip design to more advanced nodes, and continue to shrink the chip size, create higher energy efficiency, advanced AI tools to assist chip manufacturers can be faster and cheaper to achieve these shrinking process.

Filling the chip talent gap: About 2 million people worldwide will be working in the chip industry in 2022, but as the U.S., EU and China continue to push for chip self-sufficiency, the importance of advanced AI tools in chip production is growing as a way to fill the talent gap. Chip design goes through three main stages: system-level design, register transfer level (RTL) design and final physical circuit design, with advanced AI tools shining in the last stage.

The effectiveness of AI in chip design has increased dramatically

Design chips must go through physical form design, evaluation, testing, from simulation design to implementation may take years, chip design to constantly optimize the performance, power consumption, area (PPA) to minimize power consumption, increase processing speed, and produce the smallest possible chip. If you use traditional tools to optimize PPA, not only slow and labor-intensive, usually only slightly improve the PPA. using the assistance of advanced AI tools, you can find errors in placement that increase power consumption, affect performance or inefficient use of space, and then AI tools for simulation testing and improvement. AI tools can learn from previous iterations to improve PPA until the limits are reached. But what is truly revolutionary is that advanced AI can do the work autonomously and produce better PPAs than human designers using traditional EDA tools.

Elements of machine learning have been included in EDA tools for many years, but the addition of Graph Neural Networks (GNN) and Reinforcement learning (RL) techniques has dramatically increased the effectiveness of AI in chip design. Large semiconductor companies use advanced AI to develop new services for profitability. By extending their GNN and RL capabilities, semiconductor companies can not only generate their own designs, but also provide design and co-design services to their top customers to co-develop vertically specialized chips. In addition to designing chips, AI can also be applied to improve wafer failure detection and manage outsourced semiconductor assembly and test vendor networks. In addition, cloud-based EDA services can expand the potential market for advanced AI, which can be made available to smaller companies with lower technical skills and computing power as long as the EDA services are on the cloud.