Simulating Quantum Computers: A Review and Acceleration Opportunities ✛
Context
Quantum computing promises to tackle problems that are intractable for classical computers, from modeling molecules to factoring large numbers. But real quantum machines are still limited: qubits are few, and their quantum states last only microseconds. So the development and debugging of quantum algorithms and hardware go through classical simulation. This work joins the ranks of reviews on quantum circuit simulation, but with a distinct focus: it does not just list methods but systematically organizes hardware-oriented accelerations for different platforms.
Methods
The authors collected and analyzed dozens of research papers on simulating quantum circuits on CPUs, GPUs, and FPGAs. They classified simulators by level, from device physics to algorithmic level, and identified the main acceleration approaches: state vector compression, reduced numerical precision, gate clustering, circuit partitioning, memory access optimization, and parallel execution. For each approach, they provide concrete examples and achieved performance metrics.
Results
The review shows that simulation speedup can be achieved in various ways. On CPUs, state vector compression enabled simulating a 61-qubit Grover circuit on the Theta supercomputer, cutting memory requirements from 32 exabytes to 768 terabytes. On GPUs, the cuQuantum library accelerated tensor network contraction by up to 547 times compared to a baseline approach. On FPGAs, precomputed matrices and streaming processing allowed efficient complex-number operations even with limited resources. Key numbers: 61 qubits, 0.8 petabyte of memory, speedup up to 547 times.
Implications
This work stands out from neighboring reviews because it not only describes simulation methods but also systematizes hardware-dependent optimizations and highlights promising directions. Unlike studies focused on a single accelerator type or algorithm, it offers a holistic view of using CPUs, GPUs, and FPGAs. This helps developers choose the right platform and technique for a given task, and lets researchers see gaps and future opportunities.
Future development
The authors predict that specialized accelerators for quantum simulation will be scarce due to a small market. Instead, existing hardware for machine learning should be leveraged, since the key operation, matrix multiplication, is the same. They also expect increased use of FPGAs for niche tasks with non-standard data types and the development of compression methods with controlled loss of precision.
Impact
Faster simulation directly impacts the pace of quantum algorithm development and hardware verification. It helps researchers test ideas quickly without access to expensive quantum machines.
Next steps
The next step could be developing hybrid approaches that combine CPUs, GPUs, and FPGAs, as well as adapting compression and reduced-precision techniques from machine learning to quantum simulations.
Key open problems
In quantum computing, the scalability of simulation remains an open problem: exponential resource growth limits the size of simulatable circuits. This work directly links that problem to hardware capabilities and suggests mitigation paths through optimizations.
The knowledge machine's view
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The authors compiled and classified methods for simulating quantum circuits on CPUs, GPUs, and FPGAs, highlighting the main acceleration approaches: state vector compression, reduced precision, gate clustering, circuit partitioning, memory access optimization, and parallel execution. For each approach, they provide concrete examples and achieved metrics, such as a 61-qubit Grover circuit on the Theta supercomputer with 768 TB of memory and tensor network compression speedups of up to 547x with cuQuantum.
Where its strength is
Systematizing hardware-oriented optimizations across different platforms provides a holistic view and allows selecting the appropriate technique for a given task, which is especially valuable for simulator developers.
Where it could go next
- It might be interesting to see how simulation acceleration ideas apply to specific applications, such as power grid optimization, where quantum algorithms are already being tested on classical simulators.based on: Quantum First Aid for Power Grids arXiv:2607.29582v1
- Closely related is work on quantum networks, where simulating distributed quantum systems may require new approaches to parallelization and memory management; this could be a natural extension of your review.based on: Quantum Networks: How Optical Switches Link Processors arXiv:2501.05598
Worth trying
- Test state vector compression on new architectures, such as FPGAs with high-bandwidth HBM memory, for circuits with a large number of qubits.
- Consider hybrid CPU+GPU+FPGA schemes, where different parts of the simulation are distributed across platforms based on their strengths.
- Investigate how reduced-precision methods apply to simulating noisy intermediate-scale quantum circuits, where modeling errors may be comparable to hardware noise.
Why this may matter
Simulation speed directly impacts the pace of quantum algorithm development and hardware verification, and the systematization of approaches helps researchers quickly find suitable techniques without reinventing the wheel.
Nearby works in our archive
- The Noise That Helps Fix Errorssimilarity 0.58
- Shadows of Quantum States: How to Measure Faster and More Accuratelysimilarity 0.58
- Quantum First Aid for Power Gridssimilarity 0.58
- Quantum Networks: How Optical Switches Link Processorssimilarity 0.58
The closest work in the archive sits at 0.62 by meaning — the usual level for our corpus.
Simulating 30 qubits requires up to 34 gigabytes of memory, and adding just two more qubits raises the requirement to 135 gigabytes, a fourfold increase from just two qubits.
In the TV series 'Devs', a quantum computer is used to simulate reality, but its simulation on classical machines faces the same limitations described in this review.
Key numbers
- maximum number of qubits in simulation: 61
- memory for 61-qubit simulation: 0.8 petabyte
- GPU tensor network contraction speedup: up to 547 times
- memory for 30 qubits: 34 gigabytes
- memory for 32 qubits: 135 gigabytes