Pentagon launches living neural computer for drone navigation

The Pentagon’s science and technology research arm is moving forward with one of its most unusual computing programs to date, opening an industry workshop this week for O-Circuit, a research effort built around the use of living neural tissue as a computing system for future defense applications.

The Defense Advanced Research Projects Agency’s Biological Technologies Office published a special notice on April 8 ahead of the April 10 proposers workshop in Arlington, Virginia, where companies, universities, and research teams will be briefed on the planned program and invited to form teams for future bids. The initiative, formally titled Organoid Cytomorphic Intelligence Resulting from Convergent Understanding and Information Transfer, focuses on building what the agency calls biological processing units, or BPUs, as an alternative to traditional silicon-based processors.

The effort starts with a problem the Pentagon has been wrestling with for years: how to run advanced artificial intelligence systems in places where power is scarce and access to large computing infrastructure is limited. Conventional chips remain effective in fixed facilities and well-supported environments, but their power demands become a challenge for sustained use in forward locations, particularly for AI applications that require both training and real-time decision-making.

DARPA’s answer is to look beyond semiconductors and into biology.

Rather than trying to imitate the brain in software or specialized hardware, the O-Circuit program is focused on using living neurons and organoid-based neural structures as the computing medium itself. The goal is to improve the learning, inference, and memory capabilities of neural tissue systems so they can function as practical processors in military edge environments.

The idea is rooted in one clear advantage: energy efficiency. Biological neural systems process information using extremely low power compared with modern digital processors, which is one reason DARPA sees potential for field applications where battery life and endurance remain limiting factors. The agency’s draft program documents point to the natural efficiency of neural architectures as the basis for pursuing a new class of low-power computing.

The program is structured as a 42-month effort split into three phases and two main task areas.

The first track, known as Architecture, focuses on building BPUs that can learn increasingly complex tasks and retain that learning over multiple days. DARPA wants performers to move beyond short-duration laboratory demonstrations and develop organoid-based systems with more persistent memory and better adaptive behavior.

To test whether these systems are actually learning, DARPA is using a surprisingly familiar benchmark: Ms. Pac-Man.

Under Phase 1, participating teams will use a standardized Ms. Pac-Man Atari emulator provided by the government’s test and evaluation team. The game environment will serve as a common framework to measure how quickly the biological systems learn and how well they retain that learned behavior over time.

The second task area pushes the concept beyond the lab.

Known as Action, this part of the program combines the biological processor with a biological odor-detection system and a drone navigation platform. The aim is to create an integrated sense-compute-action system that can detect odor signatures and autonomously guide a drone toward the source.

DARPA’s draft documents say the system will be tested in both simulated and real-world settings, including the use of aerial and ground drones capable of carrying at least 10 pounds of payload.

One of the clearest operational goals laid out in the program is the ability to detect, identify, and autonomously locate volatilized chemical odorants.

That gives the program a direct link to defense applications beyond experimental neuroscience. In practical terms, DARPA is looking at whether a biologically based sensing and processing system can guide autonomous platforms toward specific chemical signatures in real-world conditions.

The agency has also drawn clear boundaries around the work. Approaches based on DNA computation, bacterial systems, or neuromorphic chips that merely mimic biology in silicon are not part of the program. DARPA’s focus remains squarely on living neural tissue as the core of the processor itself.

The acquisition path calls for Other Transaction prototype awards, with abstracts due by May 11, 2026, followed by oral presentations for selected teams. DARPA expects the first phase of work to begin in November 2026.

The workshop is also being opened to nontraditional contractors, academic institutions, and first-time government performers, a sign that DARPA is looking well beyond its usual defense industrial base for expertise in biotechnology, neural engineering, and artificial intelligence.

O-Circuit stands out even by DARPA standards. It brings together military AI requirements, organoid research, and autonomous systems into a single program built around the idea that future battlefield computing may rely not on smaller chips, but on living neural architectures.