The three-day conference, which began on March 12, is organised by US-based Aitomatic and the National Innovation Centre. It highlighted that AI and semiconductors are now pillars of the digital economy's future.

Notably, the two elements, "AI" and "Semiconductors" are advancing in tandem, with AI automating semiconductor manufacturing, predicting and detecting product defects, and improving production quality and efficiency.

Experts exchanging ideas on the matter

Christopher Nguyen, CEO of Aitomatic, stated that by 2030, some manufacturing plants, especially advanced production facilities, will require stricter standards, necessitating high-precision tools.

“In plasma processing, parameters such as fuel diameter, pressure, temperature, and dozens of other factors demand near-perfect accuracy. AI will help ensure these requirements are met,” he said. “AI cannot develop without semiconductors, and conversely, the semiconductor industry is evolving rapidly thanks to AI advancements. This is a symbiotic relationship where both drive each other forward.”

Anna Goldie, a senior staff research scientist at Google, sought to draw attention to the gap between software and hardware, pointing out that chip manufacturing has yet to reach its highest potential.

“While AI's computing demands are growing exponentially, hardware capabilities are not keeping pace, creating an ever-widening gap,” she said. “To unlock AI's full potential, we must shorten the chip design cycle, refine algorithms, and fully leverage data. In the future, AI will not only enhance hardware but also drive breakthroughs in various fields, from healthcare and finance to industrial production.”

Goldie introduced AlphaChip, an AI-driven method that optimises chip component layouts to reduce latency, save energy, and maximise production area. AlphaChip has already been integrated into recent generations of Google TPUs, delivering significant efficiency gains over traditional design methods.

As the world continues striving to enhance AI and semiconductor technologies, Prof. Tran Thanh Long of Warwick University in England discussed the use of memory storage and Bayesian theory, a statistical inference method where new evidence updates probability estimates, to improve AI performance and scalability.

“Bayesian theory supports AI in adjusting predictive probabilities based on new data, allowing systems to learn faster and more efficiently. This combination reduces computing resource requirements while maintaining high accuracy,” Long said.

Meanwhile, Ngan Vu of Google DeepMind introduced a research direction that proposes using circuit neural networks to create efficient logic circuit designs. By applying simulated annealing and other optimisation techniques, she and her team aim to shorten the circuit design cycle from concept to final product.

“One of the biggest challenges is balancing circuit accuracy and performance, ensuring that designs are not only precise but also resource-efficient. However, bridging the gap between AI software and hardware will unlock many new opportunities in the semiconductor industry,” Ngan said. “Applying AI to circuit design promises to revolutionise the semiconductor industry, accelerating development cycles and enabling more optimal designs.”