Quantum Neural Network Training and Inference with Low Resolution Control Electronics

TL;DR

The paper tackles the challenge of integrating quantum neural networks with cryogenic control electronics constrained by low-resolution DACs. It combines analysis of inference for pre-trained QNNs under DAC quantization with a novel training approach—temperature-controlled stochastic quantization—to overcome gradient deadlock at low resolutions. Key findings show that inference can match infinite-precision baselines with as little as 6-bit DACs, while training remains challenging below 12 bits unless stochastic updates are employed, which then enable performance equal to or better than the baseline for 4–10 bit resolutions. This work supports hardware-software co-design for scalable QML on constrained quantum hardware, suggesting substantial power and area savings in cryo-CMOS control systems as quantum devices scale.

Abstract

Scaling quantum computers requires tight integration of cryogenic control electronics with quantum processors, where Digital-to-Analog Converters (DACs) face severe power and area constraints. We investigate quantum neural network (QNN) training and inference under finite DAC resolution constraints across various DAC resolutions. Pre-trained QNNs achieve accuracy nearly indistinguishable from infinite-precision baselines when deployed on quantum systems with 6-bit DAC control electronics, exhibiting an elbow curve with diminishing returns beyond 4 bits. However, training under quantization reveals gradient deadlock below 12-bit resolution as gradient magnitudes fall below quantization step sizes. We introduce temperature-controlled stochasticity that overcomes this through probabilistic parameter updates, enabling successful training at 4-10 bit resolutions that remarkably matches or exceeds infinite-precision baseline performance. Our findings demonstrate that low-resolution control electronics need not compromise QML performance, enabling significant power and area reduction in cryogenic control systems for practical deployment as quantum hardware scales.

Figures (4)

• Figure 1: Methodology workflow.
• Figure 2: Inference accuracy of pre-trained QNN (infinite precision) as a function of D2A resolution upon deployment on a quantum computer with limited resolution DACs. Dotted line shows the test accuracy of infinite resolution QNN.
• Figure 3: Training loss vs epochs (single run) for all DAC resolutions with deterministic and stochastic ($T=1.0$) parameter update.
• Figure 4: Average train/ test accuracy vs DAC (D2A) resolution for deterministic and stochastic quantization strategies. Shaded regions show variance across 5 trials.