The human brain is often compared to a computer. Using electrical impulses, it receives inputs from the external world, integrates this information in a split second, and plans and executes responses. The brain’s immense information processing capabilities have also inspired the design of high power computing architectures and driven the development of artificial intelligence (AI) and machine learning systems.
For decades, devices like transistors and semiconductors that operate using the humble electron have dominated the computing world. However, as researchers begin to hit physical limits in how much processing power they can fit into an electronic chip, some are exploring new ways of building chips to overcome those limitations.
One solution might lie in chips that use a different signaling medium altogether: the photon particles of light, instead of the electrons of electricity. Similarly inspired by the brain’s architecture, photonic neuromorphic computing is an emerging approach that could provide unparalleled data transfer bandwidth and minimal latency compared to electronic computers, according to Bowei Dong, an A*STAR International Fellowship (AIF) scholar at Oxford University in the UK.
Ahead of his return to A*STAR, Dong delves into the challenges and opportunities in photonic neuromorphic computing and how the dynamics of light could revolutionise the next generation of computing technologies in this interview with A*STAR Research.
Q: What sparked your interest in photonics?
My first exposure to research was at the Femtosecond Dynamics lab at Nanyang Technological University, Singapore. Their work ignited my fascination with the capabilities of light. I was especially captivated by the ultrafast dynamics of light-matter interactions, which enabled the study of physics at one-quadrillionth of a second.
Over time, I’ve also seen the rapid growth of AI and its incredible capacity to assist with various human activities. Given the need for ultrafast processing in AI models, I’m convinced that photonics could excel in this area by providing ultrafast information transfer and parallelised operations. This belief has fuelled my ever-growing interest in photonics and continues to drive my passion for uncovering its full potential.
Q: How has A*STAR supported your scientific journey?
With the A*STAR Graduate Scholarship, I pursued the Integrative Sciences and Engineering PhD Programme at the National University of Singapore. This unique programme exposed me to many fields, equipping me with interdisciplinary research capabilities that I find indispensable today, especially in a field as complex and multifaceted as photonic neuromorphic computing. During my PhD training, I was also attached to A*STAR’s Institute of Microelectronics (IME), which allowed me to interact with and learn from the institute’s deep talent pool.
Following my PhD studies, I was fortunate to receive AIF’s support to pursue my passion and broaden my horizons through overseas postdoctoral research. The international setting taught me critical lessons on conducting high-quality research and helped me establish a strong global network that will be invaluable in future.
Q: How might photonics help address the limits of electronic processors?
There are innovations in electronic computing architecture that provide promising solutions, such as electronic neuromorphic computing, which essentially boosts what processors can do by modelling their architecture after the human brain. However, our work explores a fundamentally different approach to computing—using photons, not electrons, as information carriers.
Photonic computing offers several advantages over electronics, including a much larger computing bandwidth. Electronics can access a maximum bandwidth of around 50 GHz, equivalent to <1 nm in photonics, whereas the visible light spectrum alone spans 400 to 700 nm. Moreover, photonic computing experiences significantly lower latency; where data transfer in large graphical processing units can take up to microseconds per operation due to capacitive delay, photons are several-fold faster since they transmit at light speed.
Most importantly, unlike electrons that interact with each other, photons transmit independently. Each photon also bears multiple degrees of freedom (e.g., wavelength, polarisation and mode), allowing it to convey abundant information. By parallelising photonic computing across these degrees of freedom, we can derive exceptionally high computing throughput and thus address the limits of transistor scaling that Moore’s Law describes.
Photonic neuromorphic computing takes these advantages a step further; by changing both the computing architecture and the information carrier within a system, one can enable it to compute like the human brain, but at the speed of light.
Q: What tips do you have for a productive research output?
Be open to discussions, embrace interdisciplinary research and explore diverse fields. Keep thinking about how your expertise can make a difference and contribute to the work of others. It’s important to step out of your comfort zone regularly; look into different fields where one can complement or apply cross-disciplinary research. Some out-of-the-box solutions may come from people without much prior knowledge in a specific area.
It’s worth noting that publication output can vary significantly depending on the nature of one’s research. Some research focuses more on prototyping and industrial applications, with results presented to industrial partners instead of academic societies. In my case, exploring new computing paradigms requires frequent communication with the academic community to gain their feedback, which could lead to many publications.
Q: How do you tackle common challenges in your field?
Because photonic neuromorphic computing is still in its infancy, there are many technical challenges in developing its technologies, such as the on-chip integration of high power laser arrays and computing architecture optimisation. However, each challenge is also an opportunity for innovation. By focusing on a very specific challenge, I can come up with different ideas to potentially solve it, while considering what resources and approaches I can use to realise those ideas.
Moreover, such research is highly interdisciplinary; it sits at the intersection of photonics, electronics, computer sciences, mathematics and materials sciences. I often need to learn the fundamentals of other fields, enabling me to better collaborate with experts and understand their perspectives. Fortunately, working in an environment like A*STAR with its diversified talent pool provides me with access to a broad range of expertise.
That expertise is especially important as we deal with very complex systems. Problems and errors can come from anywhere and occur at any step, making troubleshooting difficult. As such, having the patience to break down complex systems into smaller components is essential. Experience also plays a crucial role, so expanding my network to consult with veteran experts is highly beneficial.
Q: What do you hope to achieve as a young researcher?
I think Singapore holds immense potential to excel as a global leader in photonic neuromorphic computing. A*STAR’s talent pool is capable of addressing various challenges related to computational architectures, algorithms and functional materials. Additionally, Advanced Micro Foundry, an A*STAR spinoff, leads the way in photonic integrated circuit prototyping.
My objective is to contribute to this mission by showcasing a prototype that outperforms existing state-of-the-art solutions. This would prove the viability of photonic neuromorphic computing as the optimal path forward, realising a paradigm shift towards next- generation computing. If successful, my long-term aspiration is to help establish a dynamic platform for this field of research in Singapore.
