Aspheric is building a new class of optical computer — a slab of lithium niobate, programmed with light, designed to be reconfigured into the photonic circuit a workload needs.
We're a pre-seed company. We're assembling the team, reproducing the underlying science, and building the IP to take programmable photonic computing out of the laboratory.
Modern inference workloads now consume hundreds to thousands of watts per GPU. Data center power draw is growing faster than the grid can keep up. Every generation of silicon delivers smaller gains for greater cost — and the physics doesn't get easier from here.
The next leap won't come from packing more transistors onto a die. It will come from changing what does the computing.
Published research has shown that projecting patterns onto a photoconductive layer above a lithium niobate waveguide can program optical neural networks directly into the chip's refractive index. No etching. No fixed wiring. No discrete components.
The thesis: a single small chip that performs a high-dimensional matrix-vector multiply at a fraction of the power of conventional silicon, and can be reprogrammed milliseconds later for an entirely different computation. Aspheric's near-term work is to reproduce that result in our own hands and improve on it.
The ambition: an FPGA for photonics — a reprogrammable substrate for the optical circuits a workload needs.
We're not selling products yet — we're building the foundations to.
We're bringing together leading researchers and engineers in thin-film lithium niobate, programmable photonics, and analog computing.
We are independently rebuilding the published academic results that this technology rests on, validating every layer of the stack in our own hands.
Working from those reproductions, we are identifying patentable improvements over the academic baseline and developing a defensible IP portfolio around the core architecture.
A waveguide simulation environment and an SDK that lets users design, compile, and deploy optical circuits onto the chip.
Targeted at specific high-value applications — starting with domains where speed, latency, and power efficiency justify a new architecture.
Moving from in-house prototyping to scalable thin-film lithium niobate fabrication.
Thin-film lithium niobate, once a laboratory curiosity, is moving to high-volume foundry production.
The AI industry's move toward low-precision inference matches the natural operating range of analog optical computing.
Federal research programs and strategic acquirers are actively funding photonic compute as the successor architecture to GPU-based inference.
Our technical foundation builds on peer-reviewed research published in Nature Physics. Aspheric is independently reproducing those results and developing improvements over the academic baseline. Our work is grounded in close collaboration with experts across the photonics community.
We're a small team. We're hiring.