Responsibilities
Define and architect high-performance, low-power NPU cores for edge AI and LLM inference applications
Evaluate and leverage open-source NPU architectures and AI acceleration frameworks
Drive hardware/software co-design for efficient execution of transformer-based models and LLM workloads
Optimize NPU architecture for
Transformer inference
KV cache handling
Attention acceleration
Memory bandwidth efficiency
Sparse computation
Quantized AI models (INT8/INT4/FP8)
Develop scalable AI compute and memory subsystem architectures
Collaborate with RTL, compiler, SDK, and AI software teams to enable end-to-end AI execution flow
Build and optimize the complete software toolchain, including
Compiler
Runtime
Kernel libraries
Model conversion tools
Performance profiling and debugging infrastructure
Enable popular AI frameworks and LLM runtimes on the NPU platform
Analyze workload bottlenecks and optimize system-level power/performance/area (PPA)
Work closely with system architects to define heterogeneous computing architectures integrating CPU, NPU, memory, and interconnect subsystems
Contribute to future AI architecture roadmap and technology strategy
Qualifications
Required
MS or PhD in Computer Engineering, Electrical Engineering, Computer Science, or related field
More than 3 years strong experience in AI accelerator, NPU, GPU, DSP, or heterogeneous compute architecture
Deep understanding of
Transformer architectures
LLM inference
Neural network acceleration
Memory hierarchy optimization
Dataflow architecture
Experience with AI model optimization techniques including
Quantization
Pruning
Sparsity
Operator fusion
Solid understanding of hardware/software co-design methodologies
Familiarity with compiler and runtime stacks for AI acceleration
Experience with RTL architecture and performance modeling
Strong programming skills in C/C++, Python, or similar languages
Preferred Qualifications
Experience with high-performance RISC-V CPU architecture or SoC design
Experience with vector processors, SIMD/SIMT architectures, or AI instruction extensions
Familiarity with MLIR, TVM, ONNX, LLVM, or related AI compiler frameworks
Experience enabling LLMs on edge or embedded AI platforms
Experience with low-power SoC architecture and mobile/edge AI optimization
Familiarity with chiplet, NoC, or advanced memory architectures
Experience with open-source AI acceleration ecosystems
What We’re Looking For
Passion for AI computing and next-generation semiconductor technologies
Strong system-level thinking and architectural vision
Ability to bridge hardware architecture and AI software ecosystems
Hands-on mindset with strong execution capability
Self-driven and collaborative team player
Why Join Us
You will have the opportunity to shape the future of edge AI computing by building cutting-edge AI acceleration technologies that enable real-world deployment of LLMs and generative AI on power-constrained devices.
For positions requiring access to technical data, Analog Devices, Inc. may have to obtain export licensing approval from the U.S. Department of Commerce - Bureau of Industry and Security and/or the U.S. Department of State - Directorate of Defense Trade Controls. As such, applicants for this position – except US Citizens, US Permanent Residents, and protected individuals as defined by 8 U.S.C. 1324b(a)(3) – may have to go through an export licensing review process.
Analog Devices is an equal opportunity employer. We foster a culture where everyone has an opportunity to succeed regardless of their race, color, religion, age, ancestry, national origin, social or ethnic origin, sex, sexual orientation, gender, gender identity, gender expression, marital status, pregnancy, parental status, disability, medical condition, genetic information, military or veteran status, union membership, and political affiliation, or any other legally protected group.
Job Req Type: Experienced
Required Travel: Yes, 10% of the time
Shift Type: 1st Shift/Days