parallel acquisition
Per complete update.
Impact
The hybrid architecture
can now compose.
Digital orchestrates. Physics carries the dense transform. The maintenance layer stays linear.
digital work
Per control cycle.
channels maintained
Independent simulation.
Route by arithmetic intensity.
Keep orchestration digital.
Read the architectureDelegate the dense transform.
Keep resident state in physics.
See photonic systemsMaintain it at the floor.
One readout and actuation frame.
See analog AIWhat this unlocks
Designs that become credible. End-to-end proof comes next.
01On-device intelligence
Photonic and analog AI coprocessors.
02Data-centre acceleration
Quantum and photonic systems.
03Autonomous field systems
RF, radar, sensing, and communications.
04Production engineering
Simulators, digital twins, and hardware in the loop.
On-device AI
The weights stay in physics.
Only vectors cross the boundary.
8 layers. 8.4 million simulated analog weights. Fidelity held under drift.
The maintenance layer is demonstrated in simulation. An end-to-end physical hybrid AI system is not yet demonstrated.
03Digital orchestration
Training, semantics, routing, planning
02Physical forward pass
n² resident transforms with n-vector input and output
01Stability contract
One parallel read, O(n) work, bounded memory per channel
Why this matters