Architect and deliver a kernel-authoring SDK and lowering tools for a novel dataflow architecture, shaping APIs and software practices while solving end-to-end hardware and software challenges.
D-Matrix jobs
d-Matrix builds memory-centric computing technology for efficient, low-latency AI inference.
34 open roles at D-Matrix
Develop and optimize software kernels for machine-learning operators on next-generation AI hardware, while improving SDK usability and performance analysis across the software toolchain.
Productize and maintain next-generation AI inference deployment software, building scalable system infrastructure and optimizing the full software toolchain in collaboration with machine learning, compiler, and hardware teams.
Develop and optimize software kernels for next-generation AI hardware, mapping machine-learning workloads and computational graphs to specialized architectures while collaborating across compiler, software, and hardware teams.
Develop and optimize software kernels for next-generation AI hardware, mapping machine-learning workloads to specialized architectures and collaborating across compiler, software, and hardware teams.
Develop and evaluate algorithms that improve large language model inference performance on DNN accelerators, balancing latency, throughput, memory efficiency, numerical precision, and model quality.
Define and lead hardware qualification strategies for AI compute platforms, validating reliability from ASIC packages to rack-scale systems and driving fixes that bridge design and production.
Analyze emerging machine-learning workloads and develop performance models and hardware-software features for current and future AI inference accelerators, collaborating across architecture, hardware, compiler, server, and kernel teams.
Lead analog-mixed-signal IC design for AI inference accelerators, owning in-memory-compute and die-to-die interface circuits from schematic through silicon bring-up while meeting low-power targets.
Lead SoC and IP design verification from architecture through tape-out and silicon bring-up, building UVM-based environments and driving debug, coverage closure, and verification sign-off.
Page 2 ยท 34 results
