Light instead of electrons
Matrix multiplication — the core operation of every neural network — is performed by interfering beams of light in silicon photonic meshes, eliminating the resistive losses that cap electronic chips.
Research
The science behind sovereign intelligence.
Peer-level research on the systems, silicon and policy that make national AI independence possible.
White PapersSovereign AINational InfrastructureAI ChipsData SovereigntyEnergy Systems
Research feature
Photonic chips: computing with light.
D-AI's frontier silicon-photonics program reimagines the AI accelerator — replacing electrons with photons to break the power and heat ceilings that limit national-scale intelligence.
10×
Compute per watt vs. leading electronic accelerators
<1 ns
On-chip matrix-multiply latency at the speed of light
~90%
Reduction in interconnect heat generation
8 λ
Wavelengths multiplexed per waveguide for parallel math
Wavelength parallelism
By multiplexing multiple wavelengths through a single waveguide, a photonic tensor core executes many independent operations simultaneously, multiplying throughput without shrinking transistors.
Energy at national scale
At gigafactory scale, a 10× efficiency gain translates into hundreds of megawatts saved — the difference between a sovereign program that is financeable and one that is not.
Sovereign fabrication path
Photonic dies can be manufactured on mature nodes rather than bleeding-edge fabs, opening a realistic route to in-country production and supply-chain independence.
Program roadmap
Phase 01In lab
Photonic tensor core
Silicon photonic mesh demonstrating optical matrix multiplication with electronic control plane.
Phase 02In design
Hybrid electro-photonic package
Co-packaged optics bonded to electronic SRAM and control logic for a complete accelerator.
Phase 03R&D roadmap
Sovereign photonic module
Rack-deployable module integrated into the LLM Box and gigafactory compute fabric.
Latest