The Defense Advanced Research Projects Agency (DARPA) has launched a program aimed at radically improving the performance of very large-scale photonic circuits.

DARPA is seeking proposals for Photonic Integrated Circuit Architectures for Scalable System Objectives (PICASSO), with an expected start date of July 1.

"Photonic integrated systems have the potential to unlock immense improvement in bandwidth, latency, energy efficiency for applications including computing, analog signal processing, and sensing," previously unreported procurement documents state.

"However, systems incorporating photonic circuits struggle to show significant system-level performance advantages over electronic systems. This is due to the photonic circuits being typically confined to narrow, isolated functions due to the limited scaling of photonic circuits."

Currently, photonics hardware is primarily used in datacom and telecom applications as transceiver circuits with tens of photonic components (such as modulators, splitters, wavelength filters, and phase shifters).

A number of photonic accelerator companies have also emerged targeting the AI space, including Lightmatter and Q.ANT.

However, DARPA claims that those approaches are limited to a "single linear mathematical operation (vector-matrix multiplication) and are capable of handling ~128-element vectors. These system architectures remain dominated by electronics."

The agency notes that the nanosecond latency of photonic circuits is "overwhelmed by the millisecond latency of the electronics," meaning that photonic accelerators still struggle to compete with GPU-based systems at the system level.

Part of the failures of the current approaches is rooted in the fundamental properties of signaling with light, DARPA notes.

The agency focused on two core technical challenges. The first is that analog optical signals in long processing chains experience significant optical attenuation and noise. The second is that signal degradation in optical systems occurs through spurious wave interactions, with performance sensitive to scattering, coupling, mode leakage, back reflections, and spurious resonance.

To get around the two problems, DARPA believes that circuit-level innovation is critical, citing the 2015 research paper Perfect optics with imperfect components.

"PICASSO seeks innovative circuit strategies for massive scaling of photonic circuits that combine multiple mitigation concepts to overcome the limitations of piecewise component-driven approaches," DARPA said.

"PICASSO will revolutionize photonic circuit architectures, from compute to light detection and ranging, by expanding the field’s focus from individual components to include a more holistic circuit- and system-level orientation, with performance measured and evaluated at the system level."

The entire program schedule is detailed in procurement documents, but key dates include phase 1 circuits being taped out at a photonics foundry by month seven, and demonstrations ready by month 18. Another 18 months later, and the mission is expected to be completed.

The program has an overall expected budget of $35 million, with a proposers' day planned on January 16. The deadline for proposals is March 6.