AI Accelerator Architect & Performance Modeler

Arago Inc.; Paris; Sur place; EUR 150 000 - 210 000; Plein temps

Avantages offerts par ce poste

Stock option plan

Healthcare coverage

Pension contributions

25 days PTO

Résumé du poste

Arago is seeking an Architecture Modeling Engineer to own performance modeling and workload analysis for its optical AI accelerator, collaborating across hardware and software teams to evaluate design choices and optimize compute kernels.

You will translate workload requirements into architecture and software requirements, and, after architecture decisions, work hands‑on with the software team to optimize workloads on the target hardware.

Qualifications

• Strong mathematical background in linear algebra, probability, numerical methods, and quantitative performance analysis.
• Strong knowledge of computer architecture, including compute pipelines, memory hierarchies, interconnects, and parallel execution.
• Strong C/C++ and Python programming skills.
• Experience analyzing and optimizing performance-critical kernels or low-level compute workloads.
• Good understanding of AI/ML workloads, tensor operations, data movement, and common execution patterns.
• Ability to work across hardware and software teams and translate workload behavior into architecture and software requirements.
• Language: English at a proficient level.
• Strong industry knowledge about AI hardware, GPUs, memory generations, rack-scale computing.
• Experience building performance models, simulators, or analytical models for CPUs, GPUs, NPUs, or other accelerators.
• Experience with profiling, benchmarking, and identifying hardware or software performance bottlenecks.
• Experience in novel computing architectures: analog computing, in-memory computing, heterogeneous systems.

Responsabilités

• Develop fast, flexible performance models to evaluate AI accelerator architecture choices and guide early design decisions.
• Work closely with the software team to characterize key AI workloads, kernels, dataflows, and system bottlenecks.
• Work closely with hardware architects to model compute, memory, interconnect, and scheduling behavior across proposed architectures.
• Drive rapid performance studies and trade-off analyses across throughput, latency, utilization, bandwidth, and efficiency.
• Help define and validate the programming model and workload mapping strategy for the accelerator.
• Following architecture definition, work hands‑on with the software team to optimize compute kernels and key workloads for the target hardware.
• Build lightweight simulation, profiling, and analysis tools that support fast iteration from architecture exploration through software optimization.

Connaissances

C/C++

Python

Mathematics

Computer architecture

AI/ML workloads

Performance analysis

English proficiency

Cross-functional collaboration

AI hardware knowledge

Performance modeling

Profiling & benchmarking

Novel computing architectures

Outils

Profiling tools

Simulators/ analytical models

Description du poste

Arago is seeking an Architecture Modeling Engineer to own performance modeling and workload analysis for its optical AI accelerator, collaborating across hardware and software teams to evaluate design choices and optimize compute kernels.

You will translate workload requirements into architecture and software requirements, and, after architecture decisions, work hands‑on with the software team to optimize workloads on the target hardware.