Digital optical computing platform for neural networks

Artificial Intelligence - humanitys new frontier - has dramatically progressed during the past decade due to the introduction of deep learning.

Deep learning applications usually require vast amounts of computing resources but microelectronics are encountering physically fundamental bottlenecks in speed, energy consumption, heating, and interconnect delay, which can no longer be resolved by scaling.

The end of Moores law urgently calls for the introduction of new devices, new integration technologies, and new architectures to efficiently accelerate deep learning applications.

Recently, optical neural networks (ONNs) based on photonic integrated circuits have shown great potential as an emerging computing architecture beyond von Neumann and Moore to overcome the bottlenecks in their microelectronics counterparts.

Yet the ONN concepts and demonstrations so far are encountering severe obstacles such as limited calculation precision, repeatability, scalability, and compatibility with microelectronics.

These obstacles are deeply rooted in the analogue computing architecture of the existing ONNs: insufficient signal-to-noise ratio due to accumulated noise and crosstalk in the system.

DOLORES aims to explore and develop a radically new digital optical computing platform to create a new class of processors, namely a digital optical neural network (DONN) processor, that is fundamentally different from any ONN scheme and solves the aforementioned obstacles.

Integrating the best of photonics and electronics, DOLORES is expected to revolutionize optical computing, and in the long-term vision enable new possibilities that are barely supported by todays computing hardware.

If successful, DOLORES can pave the way for a new era in AI, drastically accelerating neural networks and deep learning applications, which have the potential to transform many sectors of our society, including healthcare, finance, and transportation.