IEEE International Conference on Big Data Security on Cloud, IEEE International Conference on High Performance and Smart Computing, and IEEE International Conference on Intelligent Data and Security (BigDataSecurity/HPSC/IDS)
RAQET: a Reconfigurable Accelerator for QNNs Efficient Training
Samuel Farid, Taj Abdin, Hamid Reza Imani, Tarek El‐Ghazawi
Variational quantum neural networks (QNNs) require repeated forward simulations of quantum circuits during training, making state-vector simulation the dominant computational cost. This paper presents RAQET, an FPGA accelerator for the forward-pass evaluation of variational quantum circuits. RAQET offloads the full state-vector evolution to the FPGA while keeping gradient evaluation and optimization on the host processor. The design uses a deeply pipelined fixed-point datapath, a quarter-wave sin/cos lookup table, and onchip memory to efficiently process quantum gate operations. The implemented accelerator operates at 333 MHz on an AMD Kintex UltraScale+ KCU116 FPGA. Experimental results show that 16-bit fixed-point arithmetic preserves model accuracy on IRIS and MNIST datasets while improving hardware efficiency. Scalability analysis demonstrates that logic resources grow approximately linearly with the number of qubits, while BRAM usage follows the exponential growth of the statevector. RAQET maintains roughly an order-of-magnitude speedup over software-based simulation as the circuit scales with the number of qubits and ansatz repetitions.
Variational quantum neural networks (QNNs) require repeated forward simulations of quantum circuits during training, making state-vector simulation the dominant computational cost. This paper presents RAQET, an FPGA accelerator for the forward-pass evaluation of variational quantum circuits. RAQET offloads the full state-vector evolution to the FPGA while keeping gradient evaluation and optimization on the host processor. The design uses a deeply pipelined fixed-point datapath, a quarter-wave sin/cos lookup table, and onchip memory to efficiently process quantum gate operations. The implemented accelerator operates at 333 MHz on an AMD Kintex UltraScale+ KCU116 FPGA. Experimental results show that 16-bit fixed-point arithmetic preserves model accuracy on IRIS and MNIST datasets while improving hardware efficiency. Scalability analysis demonstrates that logic resources grow approximately linearly with the number of qubits, while BRAM usage follows the exponential growth of the statevector. RAQET maintains roughly an order-of-magnitude speedup over software-based simulation as the circuit scales with the number of qubits and ansatz repetitions.
Quantum algorithms such as the Quantum Fourier Transform (QFT), Grover’s Search, the Bernstein–Vazirani (BV) algorithm, and Bell-state generation are computationally intensive to emulate on classical processors due to the exponential growth of state vectors. This work presents a fixed-point, microcoded FPGA accelerator that achieves deterministic timing, near-unit fidelity, and linear scalability up to 4 qubits. The design features a microcoded control plane with pairwise and diagonal opcodes, fixed-point complex arithmetic, and subspace-gated execution (qubit0= LSB). We benchmark QFT(2–4), BV(2–4), Grover(4), and Bell(2) at 100 MHz and compare against equivalent CPU state-vector simulations. Across all algorithms, the FPGA demonstrates consistent latency reduction: QFT: 88×–154× (98.9–99.4% faster), BV: 54×–90× (98.1–98.9% faster), Bell: 282× (99.6% faster), and Grover-4: 14.2× (93.0% faster). Fidelity is near-unit for all programs except a single BV-2 run that yields 0.875 ; QFT, Bell, BV-3, BV-4 and Grover-4 achieve 1.000. Latency scales linearly with qubit count—growing by ~1.8× per additional qubit for QFT and ~1.6× for BV—confirming the deterministic and cycle-predictable nature of the microcoded architecture. Results reflect 100 MHz RTL simulation (Verilator); we report synthesis resources and post-P&R Fmax, and express speedups normalized to 100 MHz. These results show that even a modest FPGA can outperform our CPU state-vector baseline on small-n kernels while preserving deterministic, cycle-accurate timing for small-to-medium quantum workloads through spatial parallelism and hardware-level determinism.
In recent decades, Quantum Computing (QC) has gained widespread attention among researchers and industries, driven by its potential to outperform classical systems in several key areas. However, validating novel quantum algorithms poses substantial challenges due to limited qubit availability and noise in current quantum devices. Although software simulators are frequently employed to validate quantum circuits, they are time-consuming and require high memory availability since as the number of qubits increases, the memory occupation grows exponentially. Hardware emulators have recently emerged as a faster, more efficient, cost-effective alternative. This article introduces AMARETTO (quAntuM ARchitecture EmulaTion TechnOlogy), a new architecture designed to improve quantum computing emulation on resource-constrained Field-Programmable Gate Arrays (FPGAs). To provide a flexible and complete tool for quantum algorithm development, the emulator supports two universal gate sets, Clifford+T and rotational. AMARETTO mimics quantum computing’s inherent parallelism through multiple concurrent processing elements, achieving high throughput by harnessing the sparse nature of the gate matrices, coupled with pipelining in a Reduced Instruction Set Computer (RISC)-like architecture. A dedicated compiler translates OpenQASM 2.0 into AMARETTO’s instructions, enabling support for custom gates leveraging the equivalences known in the literature. Experimental results on an AMD Zynq UltraScale+ MPSoC ZCU106 demonstrate successful emulation of quantum circuits up to eighteen qubits with a throughput of12.207skgates, validated against Qiskit simulators. Our evaluation confirms competitive performance with existing FPGA-based emulators in terms of throughput and scalability, while targeting an affordable FPGA platform.
With the development of quantum algorithms and practical quantum computing devices, how to efficiently simulate emerging quantum algorithms via quantum circuit simulators on classical parallel-computing hardware platforms has raised more and more interest. The superior customized parallelism and pipeline efficiency over general-purpose processors make FPGAs ideal for large scale quantum algorithms/circuits simulation. Since the quantum Fourier transform (QFT) has been used in many quantum algorithms, such as quantum linear algebra, quantum phase estimation and quantum machine leaning, this paper presented a novel precision analysis and an efficient customized-precision QFT simulator accelerated on AMD Alveo U55C FPGAs. Customized-precision representations via high-level synthesis are employed for QFT implementation with up to 16 qubits, demonstrating up-to 8.3× speedup versus implementations in floating-point arithmetic within an error tolerance of 0.1%.
The area of quantum circuit simulation has attracted a lot of attention in recent years. However, due to the exponentially increasing computational costs, assessing and validating these models on large datasets poses significant obstacles. Despite plenty of research in quantum simulation, issues such as memory management, system adaptability, and execution efficiency remain unresolved. In this study, we introduce QEA, a state vector-based hardware accelerator that overcomes these difficulties with four key improvements: optimized memory allocation management, open PE, flexible ALU, and simplified CX swapper. To evaluate QEA's capabilities, we implemented and evaluated it on the AMD Alveo U280 board, which uses only 0.534 W of power. Experimental results show that QEA is extremely flexible, supporting a wide range of quantum circuits, has excellent fidelity, making it appropriate for standard quantum emulators, and outperforms powerful CPUs and related works up to153.16×better in terms of normalized gate speed. This study has considerable potential as a useful approach for quantum emulators in future works.
While real quantum devices have been increasingly used to conduct research focused on achieving quantum advantage or quantum utility in recent years, executing deep quantum circuits or performing quantum machine learning with large-scale data on current noisy intermediate-scale quantum devices remains challenging, making classical simulation essential for quantum machine learning research. However, such classical simulation often suffers from the cost of gradient calculations, requiring enormous memory or computational time. To address these problems, we propose a method to fuse multiple consecutive gates in each of the forward and backward paths to improve throughput by minimizing global memory accesses. As a result, we achieved approximately 20 times throughput improvement for a Hardware-Efficient Ansatz with 12 or more qubits, reaching over 30 times improvement on a mid-range consumer GPU with limited memory bandwidth. By combining our proposed method with gradient checkpointing, we drastically reduced memory usage, making it possible to train a large-scale quantum machine learning model, a 20-qubit, 1,000-layer model with 60,000 parameters, using 1,000 samples in approximately 20 minutes per epoch. This implies that we can train the model on large datasets, comprising tens of thousands of samples, like MNIST or CIFAR-10, within a realistic time frame (e.g., 20 hours per epoch). Thus, our proposed method significantly accelerates such classical simulations, making a significant contribution to advancing research in quantum machine learning and variational quantum algorithms, such as verifying algorithms on large datasets or investigating learning theories of deep quantum circuits like barren plateaus.
The exponential runtime of quantum simulators on classical machines, along with long queue times and high costs associated with real quantum devices, presents significant challenges for the efficient optimization of Variational Quantum Algorithms (VQAs) such as the Variational Quantum Eigensolver (VQE), Quantum Approximate Optimization Algorithm (QAOA), and Quantum Neural Networks (QNNs). To address these limitations, we propose a novel approach, DyPP (Dynamic Parameter Prediction), which accelerates the convergence of VQAs by leveraging regular trends in the parameter weights to update parameters efficiently. We introduce two techniques for optimal prediction performance: Naive Prediction (NaP) and Adaptive Prediction (AdaP). Through extensive experimentation and training of multiple QNN models on various datasets, we demonstrate that DyPP achieves a speedup of approximately 2.25×compared to standard training methods, while also providing improved accuracy (up to 2.3% higher) and lower loss (up to 6.1% reduction) with minimal storage and computational overhead. We further evaluate DyPP’s effectiveness in VQE for molecular ground-state energy estimation and in QAOA for graph MaxCut. Our results show that, on average, DyPP leads to a speedup of up to 3.1× for VQE and 2.91× for QAOA compared to traditional optimization, while requiring up to 3.3× fewer shots (i.e., repeated circuit executions). Even in the presence of hardware noise, DyPP outperforms existing optimization techniques, delivering up to 3.33× speedup and requiring 2.5× fewer shots, thereby enhancing the efficiency of VQAs.