DARPA launches $35M program to scale photonic circuits for AI computing
DARPA has launched a new research initiative aimed at overcoming the fundamental physical limits that have so far prevented photonic integrated circuits from delivering large-scale computing advantages for artificial intelligence and other data-intensive workloads.
The program, titled Photonic Integrated Circuit Architectures for Scalable System Objectives (PICASSO), will distribute around $35 million in funding across multiple awards to develop photonic circuits capable of predictable, scalable system-level performance. Proposals are due by March 6, with the program expected to begin in July 2026.
While photonic circuits are already widely used for optical communication, DARPA argues that their role in computation remains constrained.
Current photonic computing demonstrations are typically limited to shallow circuit depths and simple linear operations, forcing frequent conversions between optical and electronic domains.
These conversions introduce latency and power penalties that negate many of photonics’ inherent advantages in bandwidth, energy efficiency, and speed.
According to DARPA, the core challenge lies not in the lack of components, but in circuit-level architectural limitations rooted in the physics of light.
Optical attenuation and noise accumulation restrict signal fidelity as circuits scale, while spurious interference effects such as scattering, coupling, back reflections, and unwanted resonances become increasingly difficult to control across large, complex photonic systems.
PICASSO seeks to address these challenges by applying innovative circuit-level design strategies, rather than relying on new materials or devices.
Drawing parallels with electronic circuit design, where architectural techniques compensate for transistor-level limitations, DARPA aims to demonstrate that today’s photonic components can be assembled into far more capable and stable systems.
By the end of Phase 1, which runs for 18 months, program participants are expected to demonstrate photonic circuits with predictable performance at scale.
Phase 2, extending the program by a further 18 months, will focus on achieving generalised circuit functionality suitable for real computing workloads, including AI.
The initiative reflects growing interest in photonic computing as the semiconductor industry confronts power, bandwidth, and scaling constraints in electronic systems. If successful, PICASSO could mark a significant step toward photonic architectures that move beyond niche applications and into mainstream computing platforms.
