Here are our internal comparisons and the photonic glucose chip we designed in AIFIELD, as of September 2026. They show design and simulation work; they are not customer case studies or measurements on fabricated devices.
03 / Connect the system
The chip is only the beginning.
Light becomes current. Electronics turn it into a signal we can inspect.
Separate the chip from the board. Follow the detector, amplifier and converter, then switch to the heat view. This assembly study connects optical design, electronic readout and the questions a physical prototype must answer.
Spacing enlarged for inspection. Provisional PCB stack; inner copper shown as nominal planes, not exported fills. Heat is an illustration on the top mask, not a layer-by-layer simulation.
Illustrative heat distribution · no temperature scale
Touch the highlighted part. Drag empty space to turn the view.
Geometry & model
Native AIFIELD CAD: 70 × 57 mm, four copper layers, 151 component placements, 521 routes and 308 vias. The floating optical detail uses the monitor receiver die's actual generated geometry: two compact echelle spectrometers, band filter, output gratings and detector interfaces. The separated die is enlarged ×4.3 for inspection; assembling restores the physical relative size. This is a design study: provisional package pin maps, assembly and foundry assumptions remain. Material shading, electrical activity marks and the optional heat overlay are illustrative, not field, timing or thermal solutions.
A photonic glucose chip-design example
The engineering question connects an optical sensing objective to a photonic design. AIFIELD contributes the design and simulation workflow; the intended measurement belongs to ArtiSense. A design output is useful evidence of engineering activity, while physical sensing remains a separate validation task.
Internal comparisons against our own reference solver, September 2026
We will publish them with their reference, conditions and misses
Assisted workflow
In-app assistant, command line and MCP
Assistance with human review, not autonomous design
Neural prediction
Research experiments given the boundary field
Generalization and practical speed remain open
A benchmark is more than a number
A publishable comparison needs the reference tool or method, geometry, materials, boundary conditions, numerical settings, evaluation measure, runtime environment and date. It must include misses as well as successes. Accuracy and acceleration claims require that supporting evidence.
Make the next result reproducible
Our next milestone is a demonstration you can check: design inputs, relevant settings, outputs and our interpretation, shared as one package. Ask us for it.