Electronic innovation: pushing the limits of digital systems
“At that time, I picked Electrical Engineering and Computer Sciences (EECS) as my undergraduate major only because I didn’t know whether I preferred software or hardware. All I knew was that I quite liked computers and this EECS major would probably allow me to focus later,” he says, and admits that no one was more surprised than him when his application to the prestigious American university, with its close links to Silicon Valley, was successful.
A shift abroad
The transition from Hong Kong to California was challenging for a young school leaver and So admits he struggled academically and socially in the early months of his university career but he soon caught up. The experience in the late 1990s at the height of the dot com boom in Silicon Valley was both exciting and professionally invaluable. At Berkeley, he worked with his advisor, Professor Robert W Brodersen and spent most of his time at the prestigious Berkeley Wireless Research Centre, which Broderson co-founded.
He explains that he became interested in the use of FPGA-based emulators to improve the productivity of hardware system designers by allowing them to produce integrated circuits at greatly reduced time and cost. FPGA stands for field-programmable gate arrays, which are integrated circuits designed to be configured by a customer or a designer after manufacturing – hence ‘field-programmable’.
So eloquently explains that conventional integrated circuits or ASICs (application specific integrated circuits), such as the microprocessor commonly seen in home PCs, may initially cost around US$2million and take two years to fully develop, so are only cost effective in huge volumes for fixed applications.
On the other hand, developing digital systems with FPGAs is much cheaper and quicker, and is ideally suited for lower volume applications, where being first to market is paramount and where application software needs to be modified on the fly such as in space programmes.
“The Mars Rover, for example, has several FPGAs because scientists had no idea what exact functionality they would need until they got there,” he explains.
Return to Hong Kong
In 2007, on completion of his Phd at Berkeley, So felt the urge to come home to Hong Kong to continue his high technology endeavour in the place where he was born and raised. Giving up several industry and academic opportunities presented to him in the USA, he decided to join the University of Hong Kong instead, in spite of the lack of significant high-technology computer industry locally.
In 2013 he was the first recipient of the Croucher Innovation Award to research the use of reconfigurable computing technologies to improve power efficiency in warehouse scale computers. Large supercomputers and the huge data centres operated by the likes of Google and Facebook have an acute problem of massive power consumption and overheating.
“The Bay Area is still the stronghold of this science,” he says, though admits China and Japan are catching up quickly.
“Modern supercomputers produce so much heat that they can simply melt themselves if the cooling systems are not right, and almost 0.5W must be spent on heat removal for each watt of power spent on computation, adding stress to the cost and reliability of these systems,” says So.
He explains the primary motivation for Google to build some of their large-scale data centres next to rivers or glaciers is to facilitate system cooling; Microsoft has even experimented with an underwater data centre.
So’s project looks at ways to utilize reconfigurable computing technologies to replace standard CPU-based hardware which inevitably includes redundancy in terms of functionality.
“With FPGA technology, we seek to avoid doing unnecessary tasks that slow down computation and produce heat,” explains So, making it sound far more simple than the reality.
Using FPGAs for computing essentially allows the computing hardware to be re-optimized specifically for each running application. So has a natural gift for explaining highly technical concepts in an easily digestible form and he has won several teaching awards at HKU partly due to his ability to communicate complex scientific subjects to young students.
Sharing knowledge
“I do enjoy explaining diverse scientific concepts to young people,” he admits and is quite happy explaining the difference between computer science and physics or outlining the technical complexities behind the popular ‘Pokemon Go’ game.
His original intention was to join industry and make something real and he spent two years working in a high-tech start up in Hong Kong before embarking on his PhD. He retains a practical approach to his work.
“I still like to see real working systems- this, I believe, is the best form of research,” he says.
Through experience he found that blue sky research was too long-term and theoretical to engage him and commercial development too short-term and financially focused.
“Ideally, I like to look at things five to ten years down the line,” he says, which is an area that sometimes tends to fall between academia and industry and he feels strongly about being a local advocate for science and technology.
“The one thing I learned more than anything else at Berkeley is the power of diversity,” he says, and feels that Hong Kong should be less wedded to business and finance and encourage greater innovation and diversity.
“Science and technology in Hong Kong are really lagging behind now but I feel passionate about trying to address this imbalance,” he says.
Dr. Hayden So is assistant professor at the Department of Electronic Engineering at the University of Hong Kong which he joined in August 2007. He completed his BS, MS, and PhD at the EECS department of University of California, Berkeley. During his time at Berkeley, he spent most of his time at the Berkeley Wireless Research Centre where he developed the BORPH operating system for the BEE2 platform. His research mainly focuses on reconfigurable computer systems using FPGAs and Graphic Processor Units (GPUs), with a particular interest in improving energy-efficiency of the future warehouse-scale computers. He has received multiple awards for his teaching including the Outstanding Teaching Award (Team) in 2012 and the Best Teacher Award, Faculty of Engineering, University of Hong Kong 2011-2012. He was awarded the Croucher Innovation Award in 2013. His current work focuses on power-efficient heterogeneous computer clusters, hybrid FPGA-CPU computers, and their applications to medical image processing.
To view Dr So’s personal Croucher profile, please click here.