Photonic analog computing: implementation evidence
← Return to the implementation matrix
Photonic analog computing uses optical propagation and interference to perform operations, usually matrix algebra. The row concerns computation, not optical communication by itself.
| Stage | Mark | Summary |
|---|---|---|
| Reference | ● | Executable optical-network models and training methods |
| Physical | ● | Fabricated photonic neural-network chip |
| Integrated | ● | Linear and nonlinear optical operations on one chip |
| Scaled | ◐ | Multi-layer demonstration at small model scale |
| Access | — | No public programmable system located |
| Operational | — | No recurring external deployment located |
Reference — demonstrated
- Claim
- The matrix credits photonic analog computing at the Reference stage: executable optical-network models and training methods.
- Evidence
- Optical meshes and neural-network mappings can be simulated, trained, and calibrated on conventional hosts. The cited work also implements an in-situ, forward-only training method, providing an executable connection between the model and the device.
- Criticism
- Executable semantics do not establish purpose-built hardware, integration, scale, external access, recurring use, or comparative advantage.
- Sources
- Single-chip photonic deep neural network
Physical — demonstrated
- Claim
- The matrix credits photonic analog computing at the Physical stage: fabricated photonic neural-network chip.
- Evidence
- The Nature Photonics system is a fabricated photonic integrated circuit. Optical processor units physically perform the matrix operations rather than merely carrying data between electronic processors.
- Criticism
- A physical realization does not by itself establish system integration, efficient scaling, external access, recurring use, or comparative advantage.
- Sources
- Single-chip photonic deep neural network
Integrated — demonstrated
- Claim
- The matrix credits photonic analog computing at the Integrated stage: linear and nonlinear optical operations on one chip.
- Evidence
- Multiple coherent optical processor units and nonlinear activation functions are monolithically integrated on one chip. The demonstrated three-layer network computes both linear and nonlinear functions optically.
- Criticism
- A coherent system does not by itself establish efficient scaling, external access, recurring use, or comparative advantage.
- Sources
- Single-chip photonic deep neural network
Scaled — limited
- Claim
- The matrix credits photonic analog computing only partially at the Scaled stage: multi-layer demonstration at small model scale.
- Evidence
- The device integrates several stages, but the reported network has six neurons and three layers. That is sufficient to test composition, not enough to establish scaling to useful modern models once lasers, conversion, control, calibration, and thermal stability are counted.
- Criticism
- The mark is limited on the current public record: multi-layer demonstration at small model scale. Composition at the reported scale does not establish useful scaling across workloads, favorable economics, or comparative advantage.
- Sources
- Single-chip photonic deep neural network
Stages not credited
The experimental chip is not offered as an externally programmable service or appliance, and no recurring operational deployment was located.