EXCITONIC

Light, matter, and the machinesthey make possible

Newsletter
The Readout
Exciton binding energy · monolayer TMDs200–500 meV
Thermal energy kT · 300K≈26 meV
Quantum photonics market forecast$0.85B (2026) → $3.78B by 2031

Topic

Photonic Computing

17 stories on this topic, across the desks.

Nov 7

AI & SI

A Laser Neuron That Speaks in Spikes at 2 GHz — Over Ordinary Fibre

A Fudan–Zhangjiang team turns an optically injected Fabry–Pérot laser into an all-optical spike encoder that maps signal strength to firing rate above 2 GHz at 1,550 nm — fast, linearly precise, and fibre-compatible at once, with benchmark and medical results to show for it.

By Arjun Mehta ·

Nov 6

AI & SI

Slow and Wide: Olix Raises $312 Million to Wire AI Chips With Light, Not HBM

Three months after Olix's $312 million Series B valued it at $3.3 billion, the round still reads as the clearest bet yet that AI inference will be won on memory and interconnect, not raw compute — but the company has no silicon to show for it until the second half of 2027.

By Arjun Mehta ·

Nov 3

AI & SI

Printed at Record Speed: Four Million Optical Neurons in Fifteen Minutes

A Hong Kong team's parallel nanoprinting platform turns a millimeter-scale diffractive processor from a day-long fabrication job into a fifteen-minute print — and that throughput, not the accuracy alone, is what makes optical vision processors worth taking seriously as products.

By Arjun Mehta ·

Oct 30

AI & SI

Israel's Defense Ministry Buys Into All-Optical Computing

LightSolver's paid, multi-year framework agreement with Israel's defense R&D directorate is the photonic computing field's clearest customer-validation signal this year — and, notably, it discloses no contract value, no benchmark, and no named application.

By Arjun Mehta ·

Oct 29

AI & SI

The Pentagon Funds the Physics of Seeing With Light

Cornell will lead a $7.5 million DoD MURI project, 'Deep Nanophotonic Inference Systems,' to establish what 3D nanophotonic systems can fundamentally compute before electronics takes over — foundational science for hybrid optical-electronic visual inference, not a device, a benchmark, or an energy claim.

By Arjun Mehta ·

Oct 28

AI & SI

One Beam of Light, Four AI Jobs at Once

A CICT-led team reports a holographic diffractive network that runs several recognition tasks concurrently through a single phase-only spatial light modulator — no wavelength or polarization multiplexing, no moving parts, no reconfiguration — with four tasks held at roughly 90 percent accuracy, though the coverage carries no power or latency figures to judge the deployment case.

By Arjun Mehta ·

Oct 27

AI & SI

A Photonic Chip That Trains Itself — Without Backpropagation

A Shanghai team reports a photonic network that learns layer by layer in the optical domain, matching backpropagation on standard benchmarks and holding accuracy under noise and misalignment that breaks conventional optical training — but the hardware demonstrations are small, and the energy case is still unmeasured.

By Arjun Mehta ·

Oct 4

AI & SI

Neuromorphic Photonics: Spiking With Light

Brains compute with spikes, not clocks. Neuromorphic photonics tries to do the same with light — and polaritons may provide the spiking, thresholding dynamics it needs.

By Arjun Mehta ·