Improving Noisy Quantum Computation with Unitary Resynthesis

Today's quantum computers are noisy and prone to errors, limiting the breadth and depth of executable circuits. We introduce a new circuit resynthesis technique that dramatically reduces the number of one-qubit and two-qubit gates needed to realize a circuit through approximation and pattern matching, thereby improving the overall circuit fidelity.

At the core of our approach is a highly optimized approximate unitary synthesizer that decomposes arbitrary three-qubit (or larger) blocks of gates into minimal sequences of two-qubit gates. We achieve an optimal balance between the accumulation of errors due to gates and synthesis imprecision by incorporating a device-specific error model into our process. We integrate our method into the Q-CTRL transpilation pipeline and assess its benefits on various complex quantum circuits. Experimental results show up to a 50% reduction in gate counts and significant improvements across IBM Quantum devices. These results highlight our approach as a valuable tool for implementing circuits on real, error-prone quantum computers.