In Pursuit of 1000x and beyond
We build photonic AI accelerator solutions that harness light in pursuit of 1000X advances in AI compute speed and efficiency
3E8 develops lightspeed photonic AI silicon, systems and software to achieve this leap in AI compute throughput and efficiency.
Digital electronic chips are facing the sunset of Moore’s law and the acknowledged end of Dennard Scaling (energy density of shrinking transistors) which, allied to the extreme challenges and costs of shrinking transistors to below 7nm means that a new paradigm is needed to help meet the massive and growing AI compute gap over the coming years and decades.
Moreover, the physics of pushing data at higher speeds over smaller copper wires has no good solution which together with the scaling challenges result in huge energy consumption of most leading-edge digital datacenter AI chips. Such chips consume 400W+ (physical limit of single-chip packages) which requires exotic and expensive air and even water cooling solutions. The solution to this data transport problem likewise lies with using photons instead of electrons, on-chip and inter-chip. Fiber optics technology already dominates wide-area data transport for the same reasons of physics.
Now leveraging the increasingly mature silicon photonics technology driven by datacom together with novel new architectures and technologies enable us to deliver orders of magnitude advances in speed and efficiency for executing the dense linear algebra which dominates neural network computing workload.
3E8 is developing a breakthrough Photonic Tensor Core architecture which solves the hard problems of density and scalability and enables a breakthrough system frequency. The architecture is based on novel photonic memory and high-speed, low-power I/O technologies to achieve orders of magnitudes advances in speed and efficiency for massively parallel matrix multiplication and additions (MAC) which form the backbone of Convolutional Neural Networks computing.
In this mission, 3E8 partners with and builds on over a decade of cutting-edge photonic computing research at George Washington University. We have a clear runway to exploit the “unfair” advantages of photons over electrons for computation as well as transport. Among these advantages is the ability to perform multi-thread computing using dense wavelength multiplexing which promises to sustain a substantial and durable advantage over digital electronics in pursuit of 1000X and beyond.