Dizzying diminution in three decades

Professor Young chose her specialty years ago, way before the semiconductor chip enabled smartphones and other portable gadgets to proliferate.

She embarked on chip engineering at the University of Texas at Austin, which had admitted her to research into electronic design automation (EDA) at PhD level after she earned Bachelor’s and MPhil degrees at CUHK. Now a Fellow of the Institute of Electrical and Electronics Engineers, she tells CUHK in Focus about EDA, the field that quietly drives the digital foundations of contemporary society.

“Nowadays, everyone carries with them a VLSI, the small black chip inside their computer or mobile phone,” she says. “That is actually a very-large-scale integration (VLSI)”, which is an engineering miracle that has grown ever faster and smaller as technology advances.

VLSI integrates millions – now billions – of transistors, wiring networks and other electronic components onto a single chip, forming the “brain” of computers and smartphones. When Professor Young began her PhD journey in the 1990s, semiconductor chip technology measured 250 nanometres, 1 nanometre being a billionth of a metre. Today, that scale has shrunk to just 3 nanometres even as component counts soar from the millions to billions.

“A 2cm x 2cm chip now contains more elements than the population of an entire nation,” she says. “It is impossible to achieve both speed and precision in such a confined space without the right design tools.” The design of these ultra-dense circuits has long surpassed manual human manoeuvring; every stage, from logic synthesis, placement, routing, simulation to analysis and verification, all relies on advanced EDA design software.

‘Factory of 10,000 workers’

The demands of modern computing have called for increasingly complex VLSI designs. Yet most existing EDA tools in the market still run solely on central processing units (CPUs), which may lack the speed and capacity to manage today’s immense and intricate circuit designs.

To overcome these limits, Professor Young and her team are developing next-generation EDA software powered by graphics processing units (GPUs). “GPUs contain tens of thousands of parallel, specialised cores, making them very suitable to tackle the computational demands of next-generation chip design.”

The team aims to cut computation time to as little as a 20th or even a 50th of current levels. “At present, it can take days to process a medium-sized circuit. We hope to bring that down to hours or even minutes,” Professor Young says. She is under no illusion about the enormity of the challenge. “A GPU is like a vast factory of 10,000 workers – if not managed properly, it can just as easily slow things down as speed them up.”

In parallel, the team is creating a large circuit model (LCM), an artificial intelligence system purpose-built for VLSI design. In the future, by inputting functional and structural descriptions, the LCM will instantly predict circuit performance, while circuit designers can refine their design using AI.