Huawei is a leading global information and communications technology (ICT) solutions provider. Through our constant dedication to customer-centric innovation and strong partnerships, we have established leading end-to-end capabilities and strengths across the carrier networks, enterprise, consumer, and cloud computing fields. Our products and solutions have been deployed in over 170 countries serving more than one third of the world’s population.

For the Computing Systems Laboratory, we are hiring Postdoctoral Researchers in scope of an optimization-related high-performance computing platform. In the coming 1-2 years, we exclusively aim to solve foundational scientific problems related to high performance distributed- and shared-memory parallel optimization, with the goal to produce publications at top scientific magazines. On the longer term, the platform aims to solve industrial optimization problems either on-premises or as-a-Service.

We develop gradient-based inverse design for large-scale electromagnetic scattering, built on a fast multiple-scattering solver (T-matrix formalism, adjoint sensitivities) with an unusual standard of rigor: every objective is validated against an independent physical truth metric with closed energy accounting, and every design is manufacturable by construction. Our testbed, visible-band, high-NA achromatic metalenses, sits at the simulation state of the art. The methods, not the lens, are the product.

The Project

Hard scientific simulation is about to be re-platformed. The software–hardware stack built for AI, massively batched accelerator compute, differentiable programming, learned function approximation, and LLM-agent orchestration, can host physics solvers that classical HPC cannot make fast, adaptive, or exploratory enough. We have seen the early version inside this project: physics kernels rewritten as batched tensor operations (40× speedups), GPU/NPU solver paths validated to certified precision, and research campaigns executed by orchestrated AI agents under human judgment, probes designed, launched, audited, and synthesized at a cadence a human team cannot match. We are hiring the person who makes this the architecture, and who uses it to unlock capabilities that are simply unreachable today: design-space exploration at the scale of thousands of campaigns, millimetre-aperture solves overnight, learned physics surrogates that are trusted because they are truth-gated, and eventually pretrained EM-scattering priors that transfer across geometries and domains.

The project initially runs for one year, continuation depends on the project outcomes.

Responsibilities:

- Re-platforming the physics. Port the solver core to the AI stack: differentiable-programming frameworks, batched multipole algebra on accelerator fleets, and mixed precision with certified error control. The goal is not a faster port; it is a capability-class change, exact adjoints for free via autodiff, exploration parallelism as the default, and hybrid exact-physics/learned-surrogate models.

- Building better methodology. The open problems we consider most relevant are: inverse design directly in the space of physically admissible scattering operators (passivity and reciprocity as exact constraints, geometry as a projection onto the manufacturable set); safeguarding approximate forward models against optimizers that actively exploit their error (we have working instruments; the field has no theory); statistical, yield-aware design against measured process distributions; and design-space representations that scale beyond 10⁷ elements.

- Publications. Drive the methods papers: the certified-efficiency framework, the fabrication-constrained multipole bound, the objective-integrity methodology, and the re-platforming itself.

- Validation and partnership. Own the lab-validation path (designs, fabrication partner, measurement protocol, a closed sim-to-lab loop) and the commercial conversations that a certified, manufacturable design capability invites.

Transfer is the long-term goal. The formalism is domain-portable, and the re-platformed engine together with the objective-integrity is a template for any field where strong optimizers meet approximate simulators. You will pick the first transfer target and prove the method efficiency.

Your Profile

Fluent in the field (multipole/scattering theory, adjoint methods, the inverse-design literature) and genuinely at home on the AI stack: you have moved a real scientific workload onto accelerators or differentiable frameworks and know where the numerical bodies are buried (precision, reduction order, conservation laws). Strong in both C++ and scientific Python. Burned by a metric at least once, and you changed how you work because of it. Comfortable directing agentic AI research workflows; that leverage is part of what you inherit. Balanced by temperament: a clean theorem, a passing hardware measurement, and a 40× kernel speedup should all feel like wins.

Qualifications:

- PhD in physics, applied mathematics, electrical engineering, computer science, or a related technical field (or MSc with equivalent research experience).

- Solid numerical linear algebra, at both the theory and tooling level: BLAS/LAPACK ecosystems (MKL, cuBLAS/cuSOLVER or equivalent), dense and sparse factorizations, SVD/eigensolvers, conditioning and error analysis, mixed-precision arithmetic.

- HPC experience: MPI and OpenMP, SLURM-managed clusters, performance engineering and profiling (e.g., Nsight, VTune, roofline analysis), vectorization.

- AI-stack proficiency: PyTorch and/or JAX; custom accelerator kernels (CUDA, Triton) a plus.

- Scientific software engineering discipline: modern C++ and Python, version control, CI, and regression/golden testing of numerical code

- A publication record in computational physics, photonics, numerical methods, or a related area.

- Desirable: computational electromagnetics tools (FDTD, RCWA, FEM), nanofabrication literacy (EBL/RIE process constraints), and experience with ML surrogates for physical operators.

What We Offer:

· Competitive salary and benefits package.

· Opportunities for professional growth and development.

· Be part of innovative projects that make a difference.

· Access to state-of-the-art technology and tools.