DARPA Launches HARQ Program to Build Heterogeneous Quantum Architectures

April 14, 2026 – The Defense Advanced Research Projects Agency (DARPA) has launched the Heterogeneous Architectures for Quantum (HARQ) program, a 24-month initiative to develop quantum computing systems that combine different qubit technologies into unified architectures. Nineteen teams from 15 organizations have been selected across two parallel workstreams.

The program addresses what DARPA Program Manager Justin Cohen calls the field’s “one-qubit-to-rule-them-all” mindset. Today’s quantum computers are built around a single qubit type, forcing every function (processing, memory, communication) to rely on that one technology’s strengths and weaknesses. HARQ takes inspiration from classical computing, where CPUs, GPUs, and specialized accelerators each handle the tasks they’re best suited for.

The bottom line: HARQ signals that the U.S. defense establishment considers heterogeneous quantum architectures, systems combining multiple qubit types connected by quantum interconnects, a credible path to scaling quantum computing beyond the limitations of any single technology. For anyone tracking the path to a cryptographically relevant quantum computer, this is relevant: if heterogeneous architectures prove more efficient at scaling than monolithic ones, they could compress the timeline to useful quantum computation, including cryptanalysis.

Two Workstreams

HARQ is organized into two complementary tracks:

MOSAIC (Multi-qubit Optimized Software Architecture through Interconnected Compilation) focuses on software frameworks and circuit compilers that can optimize a quantum algorithm’s performance by distributing tasks across different qubit types. The goal is to produce compiled “mosaics” of physical circuits that outperform anything a single-platform compiler could generate. Selected teams include Infleqtion, MemQ, Q-CTRL, the University of Michigan, and the University of Pennsylvania.

QSB (Quantum Shared Backbone) tackles the hardware challenge: building high-fidelity interconnects that enable communication between different qubit species. This means engineering the physical interfaces that let a trapped-ion processor exchange quantum states with a superconducting processor, or a neutral-atom system talk to a photonic one, without destroying the quantum information in transit. Selected QSB teams include IonQ, Harvard University, Stanford University, UC Berkeley, the Australian National University, Carnegie Mellon University, and École Polytechnique Fédérale de Lausanne (EPFL).

My Analysis

The Classical Computing Parallel

I’ve been writing about heterogeneous quantum computing architectures for some time, and HARQ represents the most concrete government validation of this approach to date.

The classical computing analogy is instructive. No one builds a modern data center with only one type of processor. CPUs handle sequential logic. GPUs accelerate parallel computation. TPUs and other accelerators optimize specific workloads. The heterogeneous model won because different computational problems have different resource profiles, and specialized hardware handles each profile more efficiently than any general-purpose design.

Quantum computing faces a similar reality. Each qubit modality brings distinct advantages. Superconducting qubits (IBM, Google) offer fast gate speeds. Trapped ions (IonQ, Quantinuum) deliver the highest gate fidelities and longest coherence times. Neutral atoms (QuEra, Atom Computing) can be arranged in large, flexible arrays. Photonic qubits (PsiQuantum, Xanadu) operate at room temperature and interface naturally with fiber networks. No single technology leads in every dimension simultaneously.

A heterogeneous quantum computer could, in theory, assign each function to the qubit type best suited for it: trapped ions for high-fidelity processing, neutral atoms for memory, photonic qubits for communication between modules. The challenge, and this is what QSB must solve, is building the interconnects that translate quantum states between these different physical systems without unacceptable fidelity loss.

The Interconnect Problem Is the Hard Part

MOSAIC is important work, but QSB is where the physics gets difficult. Converting a quantum state from a microwave-frequency superconducting qubit to an optical-frequency photon (for fiber transmission) and then to a trapped-ion qubit requires transduction across wildly different physical regimes. Each conversion step introduces noise and potential decoherence. Making this work with the fidelity that fault-tolerant quantum computing demands is one of the hardest open problems in quantum engineering.

IonQ’s selection for QSB is notable. The company announced it will contribute quantum memory technology based on synthetic diamond, aiming for high-fidelity communication between diverse qubit species. IonQ also recently demonstrated what it calls the first photonic interconnection of two commercial trapped-ion quantum systems, done in collaboration with the Air Force Research Laboratory. That’s a direct proof point for the kind of heterogeneous connectivity HARQ envisions.

Connections to the CRQC Capability Framework

In my CRQC Capability Framework, heterogeneous architectures touch several capability dimensions simultaneously. The interconnect quality maps to qubit connectivity and routing (B.4). The ability to maintain coherence across modality boundaries affects continuous operation stability (D.3). And the manufacturing and integration challenges of multi-modality systems are squarely within engineering scale and manufacturability (E.1).

If HARQ’s 24-month timeline produces working prototypes of heterogeneous quantum interconnects, that would represent meaningful progress on dimensions B.4 and E.1 — the connectivity and engineering layers that are often underappreciated in headline-driven qubit-count reporting.

What to Watch

The 24-month program timeline puts initial results in early-to-mid 2028. That aligns with the broader fault-tolerant quantum computing timeline: IBM targets Starling (its first large-scale fault-tolerant system) for 2029, and the IBM-Cisco collaboration aims for networked fault-tolerant proof-of-concept around 2030. HARQ’s interconnect work could feed directly into these industry timelines.

Watch for the first cross-modality quantum gate demonstrations coming out of QSB. The fidelity numbers will tell us whether heterogeneous quantum computing is a near-term engineering program or a longer-term research aspiration. The software work in MOSAIC will be important too — but compilers can always be improved incrementally. The physics of high-fidelity quantum transduction either works or it doesn’t.