Electronic devices – from our mobile phones to personal computers, even the satellites in outer space – are integral to modern life. As the demand for more powerful machines grew, electronic systems design has become increasingly complex, driving rapid development in integrated circuit (IC). Very large-scale integrated circuits (VLSI) nowadays can include over a billion transistors, making their design impossible without the help of Electronic Design Automation (EDA) software and tools. Professor Liu Jinwei, ssistant Professor at the Department of Computer Science, shares his insights in this constantly evolving field.
After earning his Bachelor’s degree in Computer Science and Technology from Sichuan University, China, in 2018, Professor Liu began pursuing his PhD in Computer Science and Engineering at The Chinese University of Hong Kong, which he completed in 2022. Introduced to EDA by his doctoral supervisor, Professor Liu felt that the discipline mainly involved optimisation-related issues that align well with his academic background, and became attracted by the unique opportunities presented by this field.
Specifically, Professor Liu has been focusing on VLSI routing, which he describes as one of the most challenging aspects of physical design in electronic systems. “Routing in the context of EDA refers to the process of interconnecting circuit components using metal wires to realise the functionality of a circuit design. This is an NP-hard problem, as the number of pins increases, the time needed to solve it will grow exponentially. The layouts of multiple interconnects must be meticulously coordinated to avoid overlaps or other types of conflicts,” he notes.
Professor Liu’s groundbreaking research has already generated many useful and efficient routing algorithms and methodologies, delivering state-of-the-art performance with respect to both the quality of results and routing speed. His team is in the process of developing a detailed router, which has already shown promising results and out-performed the current best academic solution with respect to most of the performance metrics.
Having been involved in EDA research for several years, Professor Liu and his team had won several international accolades, including the ISPD 2020 contest, organised by the prestigious ACM International Symposium of Physical Design.
“This contest was especially memorable because the challenge was to deploy deep neural networks (DNNs) on the wafer-scale engine (WSE) developed by Cerebras Systems Inc., which contrasts with traditional approaches by utilising an entire wafer for a single, massive circuit. It offered unparalleled scale and parallel computing capacity,” Professor Liu recalls.
The team initially developed several complex solutions, but the result was not satisfactory. Therefore, they took a more simplistic approach – by optimising kernel shapes and placement based on data flow, yielding efficient use of resources. Subsequently, their algorithms were extended to map finite element models – commonly used in engineering for solving partial differential equations – onto the WSE.
Despite its potential to optimise the performance of different kinds of electronic devices, Professor Liu believes EDA currently faces two critical shortcomings.
The primary challenge is scalability and efficiency. With constant advancements in hardware technology, the complexity of its components is also growing. As described in Moore’s Law, the number of transistors on a microchip increases exponentially. When it reaches a certain level, the original algorithms become less efficient, therefore necessitating an update.
For example, Graphic Processing Unit (GPU) processing speeds can be improved by distributing workloads across multiple threads to enable parallel execution. However, as chip designs evolve, previously effective algorithms may no longer suffice. To run the design flow just once may take several days, with debugging extending to months. Reduction of these timeframes would help to accelerate issue identification and lower semiconductor production costs.
Another key issue is that ‘automation’ remains limited within EDA. While algorithms may address specific problems, real-world applications frequently present variations requiring substantial engineering intervention. Achieving full automation remains challenging, prompting efforts to apply artificial intelligence (AI) and machine learning techniques that enable algorithms to adapt autonomously to new scenarios.
Coincidentally, the integration of AI and machine learning into EDA processes has become a prominent direction for research laboratories worldwide.
“In hardware design, for instance, client specifications regarding chip functions can be directly transformed by AI into hardware description language (HDL) code representing circuit structure and behaviour. This process utilises generative models to generate HDL according to the desired functionality expressed in natural language. Other academics employ AI for predictive analysis, using neural networks to anticipate outcomes in advance and enable pre-emptive identification and rectification of design issues. Again, this overcomes the time-consuming process for developing the workflow and debugging,” Professor Liu shares his observations.
Professor Liu finds working in EDA particularly rewarding due to the field’s relatively modest size, which enables rapid translation of research results into practical applications. Recently, an algorithm he had published was swiftly implemented in industry, underscoring the tangible impact of his work.
“Furthermore, I am grateful for the robust institutional infrastructure at HKBU, including state-of-the-art CPUs and GPUs. These facilities significantly streamline my research progress, saving the hassles of equipment procurement with individual funding,” he states.
The experience in ISPD 2020 contest stemmed Professor Liu’s belief that “Less is More” when it comes to creating suitable solutions. “When we tackle difficult challenges, it is often easier to provide complex solutions, creating an individual answer for each different scenario. However, the solution may be clumsy and inefficient. If you examine and compare the scenarios, you may find the similarities and linkages to support a neat and unified approach. While this is more difficult, I believe it is invaluable,” he shares.
For students who are considering a research career, Professor Liu encourages them to maintain a broad perspective: “The world presents diverse and intricate problems. Do not limit yourself to popular topics. While they may attract attention, they are not necessarily suited to every researcher, nor do they guarantee success. By exploring varied directions, students can discover paths that align with their interests and strengths, maximising their potential to make meaningful contributions.”
