About the Role
As an ML Research Engineer at SEMRON, you will design the algorithms and quantization schemes that unlock efficient, high-accuracy inference on our analog in-memory compute platform. Your work will bridge cutting-edge quantization research, mathematical modeling, and hardware-aware algorithm design, ensuring that deep neural networks execute with maximal accuracy and throughput on our custom silicon.What you will do:
- Research and develop novel analog-aware quantization methods (PTQ and QAT) tailored to in-memory compute constraints
- Design mathematically principled matrix-vector multiplication algorithms that exploit sparsity, noise resilience, and non-idealities to improve hardware efficiency
Collaborate with analog hardware engineers to define algorithmic requirements and guide co-development of compute primitives
What you should bring in:
- PhD or equivalent research experience in machine learning, applied mathematics, or a related field
- Strong understanding of quantization, model optimization, and numerical methods for DNNs
- Proficiency in Python and PyTorch, with the ability to rapidly prototype and evaluate research ideas
- A research mindset: curiosity, rigor, and the ability to explore and discard ideas efficiently
Helpful but not required:
- Contributions to quantization libraries or novel compression methods
- Publications in top-tier ML venues (NeurIPS, ICLR, ICML, etc.)
- Familiarity with analog computation challenges (noise, nonlinearity, limited precision, etc.) and the ability to abstract them into robust algorithms
- Experience collaborating with hardware teams or formulating algorithm-hardware co-design strategies