The Signal
Pandora Compiles a 7.7-Billion-Gate Quantum Circuit
The Pandora software compiled a 100-by-100 Fermi–Hubbard instance into roughly 7.7 billion Clifford+T gates, according to its authors. Their paper also describes how the tool stores and rewrites circuits and reports tests of multithreading,
Quantum.world Team · Written with AI · fact-checked ·
A quantum-circuit tool called Pandora was used to decompose a 100-by-100 Fermi–Hubbard instance into roughly 7.7 billion Clifford+T gates, its authors report in a paper submitted on August 7, 2025. Quantum software must compile and manipulate large circuits before they can be analyzed or prepared for execution on hardware. The paper describes Pandora as a tool for storing, rewriting and partitioning circuits, as well as for full compilations and equivalence checking.
A database engine for large circuits
Authors Ioana Moflic and Alexandru Paler implemented Pandora on top of PostgreSQL. The software represents circuits as doubly linked lists and stores them in database tables. Its rewrite engine uses stored procedures—sets of database queries executed as transactions—and is accessed through a Python API. The paper says the tool uses Python interfaces with Qualtran, pyLIQTR, Qiskit and TKET.
The authors describe separate threads for rewriting circuits and for inserting or extracting them. Rewrite threads can search for gate patterns and apply matching rewrite rules in parallel, with database locks used to coordinate concurrent changes. To limit memory requirements during import and export, Pandora processes a fixed-size batch of gates at a time and frees the memory used by each batch before moving to the next.
The paper also describes a map-reduce setup for distributing tasks across nodes in a high-performance computing system, with each node running a Pandora container. The authors write:
“Pandora can easily handle billions of gates and can stream circuit partitions in resource estimation pipelines at very high rates.”
What the large-circuit tests show
The paper identifies the 100-by-100 Fermi–Hubbard instance, obtained from pyLIQTR, as one of the largest utility-scale circuits decomposed with Pandora. It contained on the order of 7.7 billion Clifford+T gates. The authors also report a full compilation of a 1,024-bit Shor’s algorithm circuit using Qualtran. They note that, because Qualtran did not include a lower-level decomposition of one component, CtrlScaleModAdd, they used a decomposition from pyLIQTR.
A multithreading test used a 50-by-50 Fermi–Hubbard circuit generated by pyLIQTR. The resulting 514 million Clifford+T gates were stored in Pandora at speeds of up to 270,000 gates per second using 16 cores. That is a reported storage speed, not a measure of the time needed for a complete compilation. The paper separately reports measurements for compiling and partitioning circuits.
Benchmarks have a defined scope
For comparisons with Qiskit and TKET, the authors generated random circuits containing between 10,000 and 100,000 CNOT gates and tested a particular rewrite operation. The paper reports that Pandora performed better on circuits with more than 10,000 CNOT gates when the targeted rewrite was applied infrequently, attributing the advantage to faster pattern searches. These results concern the tested rewrite task, not every workload handled by the other compilers.
For partitioning tests, the paper describes a pipeline in which circuits are decomposed into Clifford+T gates with pyLIQTR, stored in Pandora, partitioned, and then extracted. The authors report that, in this pipeline, partitioning and extraction were more efficient stages than transpilation and insertion. Their results measure distinct parts of a classical software process; they do not report execution of the billion-gate circuit on a quantum processor.