Using AI to design microchips
Artificial intelligence (AI) is on everyone’s mind right now. With the rise of ChatGPT and other AI software expanding our potential, every industry is wondering how AI can help them. The electronics industry will not miss out.
One company providing industry insights, Deloitte Global, predicted this year semiconductor companies will spend around $300 million on AI tools.
Granted, in the grand scheme of things $300 million is not a huge amount compared to the entire market, worth $660 billion. However, the return on investment is huge and can’t be ignored.
But staff should not fear, these tools are used to help, not replace, engineers. Chip design tools have been created by companies specialising in Electronic Design Automation (EDA). The tools are usually to help engineers design and simulate chips, without the need to physically manufacture them.
The price of the future?
These AI tools aren’t for everyone – a single license could be very pricey, and well above what smaller companies could afford. This would be a small price to pay for those who can afford it though, since the resulting designs could be worth billions.
It is also possible for companies to create their own AI tools in-house instead of buying from an EDA company. This, however, would need the company to have AI expertise already.
The great thing about working alongside AI is it greatly improves efficiency and size of semiconductors. AI tools can design chips under the 10nm process node to make them even smaller and more efficient.
Another advantage of using AI currently is to bridge the employment and skill gap. Because of legislation like the US and EU Chips Act, there’s a need for many more highly-qualified and skilled people within the semiconductor industry. But filling those new jobs does not happen instantly, in fact it could take years to fully train people to fill those roles. In this case, using AI in the meantime makes perfect sense, giving current engineers room to breathe.
AI already has some sway in the industry. Approximately 30% of semiconductor device makers surveyed by McKinsey said they were already generating value through AI or ML. The other 70% are still only in the starting stages of implementing the technology.
A learning curve
Within the umbrella term of AI, there are technologies that are used including graph neural networks (GNNs) and reinforcement learning (RL). RL is the repetitive running of simulations and finding a positive result through trial and error. AI can run these simulations at such a high speed, and without the use of a physical version of the electronic components.
GNNs, on the other hand, are advanced in other ways. This machine learning algorithm analyses graphs made up of nodes and edges, extracting information and making predictions. Because the structure of a chip share a similar structure to these graphs, GNNs can be used to analyse and optimise chip structure.
You don’t need machine learning to learn all about how Cyclops Excess can help you. With a range of selling options tailored to fit you, and an extensive stocklist with everything you need, Cyclops Excess has it all. Contact us today at email@example.com.
- 0 Comment