Why photovoltaics is a quantum-native problem
Designing a better solar absorber is, at its core, a problem of solving the electronic structure of a strongly correlated material. Classical density-functional theory (DFT) handles single-reference systems well enough — silicon, simple III-V semiconductors — but the materials that matter for next-gen solar are exactly the ones DFT struggles with. Hybrid organic-inorganic perovskites, lead-free double perovskites, kesterites, and chalcogenide thin films all exhibit multi-reference character: lone-pair stereochemistry on the B-site cation, spin-orbit coupling at heavy halides, polaronic distortion, and defect states deep in the gap. Standard DFT functionals smear these out. Hybrid functionals like HSE06 patch some of it. GW + Bethe-Salpeter gets closer but scales as O(N^6) and runs out of memory before you reach a useful supercell.
Quantum photovoltaics — meaning electronic-structure simulation of solar absorbers on a quantum processor — sidesteps that scaling. A fault-tolerant quantum computer can in principle solve the full configuration interaction problem in polynomial time. We are not at fault tolerance yet. But variational and quantum-subspace methods on a 100-qubit superconducting machine are already at the threshold of telling us things classical methods cannot, particularly for active-space embeddings inside a perovskite unit cell.
The active-space problem on a 100-qubit machine
You do not put an entire perovskite supercell on the QPU. You partition it. The standard pattern is DMET or CASSCF embedding: classical DFT handles the bulk, a quantum solver handles the strongly correlated active space. For methylammonium lead iodide (MAPbI3), the active space that matters is the Pb 6s / 6p and I 5p manifold around the band edges — roughly 20 to 40 spin-orbitals once you include relativistic corrections. Map that with a Jordan-Wigner or Bravyi-Kitaev encoding and you are in the 40-to-80 logical qubit range.
On a 100-physical-qubit transmon device with heavy-hex connectivity, you do not have logical qubits yet — you have noisy physical ones. Which means the realistic 2027 workload is:
- Active spaces of 12-20 spin-orbitals for serious benchmarking
- Variational Quantum Eigensolver (VQE) with hardware-efficient or UCCSD ansätze
- Quantum Subspace Expansion (QSE) for excited states — directly relevant to optical gaps
- Sampling-based Quantum Diagonalization (SQD) for moderately larger problems
- Probabilistic error cancellation and zero-noise extrapolation as the error-mitigation layer
This is not full configuration interaction. It is, however, a reproducible quantum benchmark on a class of materials where classical methods disagree with each other by more than 0.3 eV on the fundamental gap.
What we want to compute, and why
The defect physics of perovskites is the commercially relevant question. Champion-cell efficiency is now over 26% for single-junction perovskite, but operational lifetime and lead toxicity remain the blockers to deployment. Both reduce to defect chemistry:
- Iodide vacancy migration — the VI+ formation energy and migration barrier dictate hysteresis and ion-drift degradation. Quantum chemistry on the active region around the vacancy gives you the answer DFT-PBE gets wrong by hundreds of meV.
- Lead substitution — Sn, Ge, Bi, Sb double perovskites all need accurate treatment of the lone-pair s2 chemistry. This is exactly where multi-reference quantum solvers earn their keep.
- Excitonic binding and hot-carrier lifetimes — relevant for breaking the Shockley-Queisser limit. QSE and quantum equation-of-motion methods give you excited-state energies on the same footing as the ground state.
- Interface chemistry — the perovskite/transport-layer interface dominates real-device efficiency. Embedding lets you put the interface atoms in the active space.
None of this replaces experimental synthesis. It re-orders the candidate list. If you are screening a hundred A-site/B-site/X-site combinations for a Sn-Pb mixed perovskite, getting the band-gap ordering right before you spin-coat anything is what saves the lab eighteen months.
Toolchain and developer surface
The Tipperary machine exposes the standard ecosystem. Workloads come in as OpenQASM 3 circuits, compiled from Qiskit, PennyLane, or Cirq. For chemistry specifically, the relevant stack is:
- PySCF for the classical reference, mean-field, and integral generation
- OpenFermion or Qiskit Nature for fermion-to-qubit mapping
- Tangelo, QForte, or in-house DMET drivers for embedding
- Native gate set:
√X,X,RZ,ECRon the heavy-hex lattice — same instruction-level abstraction researchers already write against on IBM-style hardware
The transpiler matters more than the algorithm. A naïve UCCSD circuit on 20 qubits compiles to tens of thousands of two-qubit gates, which is hopeless given current ECR fidelities. The work is in the ansatz: ADAPT-VQE, qubit-ADAPT, and hardware-efficient ansätze with symmetry preservation cut circuit depth by an order of magnitude. We will publish reference circuits and noise-extrapolated benchmark results for a small set of canonical perovskite active spaces once the device is past first-light multi-qubit.
Hardware reality and timeline
The system is 100 physical qubits, fixed-frequency transmons in heavy-hex, operating below 15 mK in a dilution refrigerator. That is a near-term intermediate-scale quantum (NISQ) machine. It is not error-corrected. The surface-code roadmap exists but a logical qubit at useful distance costs roughly a thousand physical qubits — a future generation, not this one.
The honest delivery schedule for photovoltaic workloads:
-
Research collaboration or early access
Direct with Michael. No charge for the call.
Book a research call →