HPC-inspired Quantum Middle Layer Supports Diverse Technologies, Enabling Backend-Neutral Execution with Typed Data

Quantum computing promises revolutionary advances, but realising this potential requires software that can adapt to diverse and rapidly evolving hardware. Stefano Markidis, Gilbert Netzer, and Luca Pennati, along with Ivy Peng, all from KTH Royal Institute of Technology, present a new blueprint for a ‘quantum middle layer’ designed to bridge this gap. Their approach draws inspiration from high-performance computing, creating a system where programmers define what they want to compute, rather than how it should be done on a specific machine. This technology-agnostic design allows the same program to run on different quantum platforms, such as gate-model simulators and quantum annealers, simply by changing a separate ‘context descriptor’, representing a significant step towards portable and scalable quantum software.

Abstract Interfaces for Hybrid Quantum Computing

This research introduces a design for a middleware layer that facilitates interoperability and portability in hybrid quantum-classical computing. The core idea centers on defining backend-neutral contracts, abstract interfaces that decouple algorithmic descriptions from the specifics of underlying quantum and classical hardware. This allows developers to write code that can run on diverse platforms without modification, promoting code reuse and simplifying the development process. The proposed middleware separates the algorithmic description from the details of the quantum and classical hardware, allowing for flexibility and adaptability as new hardware emerges.

By defining standard contracts, the middleware enables different quantum and classical computing components to work together seamlessly. This design is intended to be adaptable to evolving hardware and control models, including gate, pulse, continuous-variable, and annealing approaches, without breaking existing code. This paper emphasizes the design principles for such a middleware rather than presenting a fully implemented system.

Technology-Agnostic Quantum Program Specification and Execution

Scientists engineered a novel middle layer designed to support quantum applications across diverse technologies, drawing inspiration from high-performance computing (HPC) libraries and middleware. This system achieves technology-agnosticism by separating the intent of a quantum program from its specific implementation on a given hardware platform. Programs define operations using typed data and operator descriptors, specifying what transformations are required without committing to specific gates, pulses, or hardware details. These intent artifacts remain separate from a context descriptor, which details execution parameters and can be altered per backend without modifying the core program logic.

The team demonstrated this portability with a proof-of-concept implementation utilizing JSON files for descriptors and two backends: the IBM Qiskit Aer simulator for gate-model quantum computing and a simulated D-Wave Ocean system for quantum annealing. On a Max-Cut problem, the same program successfully ran on both backends by simply varying the operator formulation and adjusting the context descriptor, demonstrating the adaptability central to the design. Researchers characterized the system’s core components, including quantum data type descriptors that assign explicit meaning to quantum registers, and quantum operator descriptors that express logical transformations independent of their physical realization. A context descriptor manages the execution policy, allowing control over parameters like error correction or multi-quantum processing unit execution, while context services provide supporting functionality. Algorithmic libraries consume these data types and operators, packaging them for submission to a backend. This modular approach, mirroring principles found in HPC libraries like BLAS and MPI, prioritizes composability and late binding, enabling the system to evolve alongside advancements in quantum hardware.

Unified Quantum Programs Across Diverse Hardware

Scientists have developed a novel middle layer blueprint designed to support quantum applications across diverse technologies, including gate-based systems, quantum annealers, and continuous-variable systems. This work addresses limitations in current quantum programming approaches by introducing explicit data and operator descriptors, enabling a single program to run on different hardware platforms without modification. The core achievement lies in separating the intent of a quantum computation from the specific implementation details, allowing for late binding of parameters and adaptive control. The team demonstrated this portability with a proof-of-concept implementation utilizing JSON files for descriptors and two backends: the IBM Qiskit Aer simulator and the D-Wave Ocean simulator.

On a Max-Cut problem instance, the same typed problem successfully executed on both backends simply by varying the operator formulation and context. This flexibility is achieved by defining what registers mean and which logical transformations are required, independent of specific gates, pulses, or backend characteristics. Researchers emphasize the importance of explicit data typing, noting that current systems often lack information about how data is encoded in quantum registers, hindering composability. The new middle layer requires unambiguous decoding rules, including bit or mode ordering and datatype interpretation, ensuring that independently defined components can interoperate seamlessly. Furthermore, the system allows for explicit control over the execution context, such as the use of logical qubits for quantum error correction or specific settings for quantum annealers. The resulting architecture prioritizes minimalism, composability, and late binding, mirroring successful design principles from High-Performance Computing (HPC) middle layers like BLAS and MPI.

Portable Quantum Programs Across Diverse Platforms

This research presents a blueprint for a new middle layer designed to support applications across diverse quantum technologies. Inspired by established high-performance computing practices, the system separates the definition of a problem’s intent from the details of its execution on specific hardware. Programs specify what transformations are required using typed data and operator descriptors, without committing to particular gate sets, pulse sequences, or backend implementations. This approach allows the same problem to be executed on different quantum platforms simply by changing the execution context and operator formulation.

The authors acknowledge that the rapidly evolving nature of the quantum computing landscape precludes the establishment of rigid standards at this stage. Consequently, their work prioritises backend-neutral contracts, allowing for interoperability between ecosystems while accommodating new hardware features and execution policies. Future work could involve implementing a surface language and compiler to generate a portable intermediate representation, potentially using frameworks like MLIR or Protocol Buffers, to further enhance the system’s flexibility and efficiency.