Mike Kraus Email & Phone Number

Naperville, Il, Us

Mountain View, California, United States

Technical lead for the customer hardware (inference accelerator) evaluation Early Access Program(EAP). Work directly with customers to understand their AI workload needs and help them take full advantage of Lightmatter’s photonic technology for inference acceleration, delivering 99% fp32 accuracy using analog compute.Prototype, profile, and optimize computer vision, autonomous driving and transformer workloads for LM hardware.

Palo Alto, California, United States

Created and optimized client-specific solutions for Deep Learning (DL) workloads using an IPU strategy, delivering a customized end-product that maximized efficiency. Worked with software engineering to migrate an open-source quantum chemistry model to an IPU-based graph neural network (GNN). This was the first customer deployed IPU GNN, and reduced customer time to train from days to hours.Participated actively with the sales division to identify and engage potential clients, showcasing how they can leverage technical advantages from the IPU, ensuring familiarity with application capabilities, and driving sales.Developed comprehensive training documents and mentored newcomers on the team, streamlining the onboarding process and fostering a positive learning environment.Created reproducible, independent test cases that helped the software team identify distributed model training bottlenecks, to resolve scaling performance issues.

Wilsonville, Oregon, United States

Worked to formulate, develop, and integrate two custom DL models into Calibre’s computational lithography modeling tool, nmModelFlow. This improved the accuracy of existing customer models by fitting non-linear errors. Managed a multidisciplinary team as the beta product release owner.Migrated all production DL training and inference Tensorflow code from the 1.x to 2.x framework.Owned the technical integration of Python-based DL models into Calibre’s C++ code base.Developed a coarse, image-based, technique for creating computational kernels to reduce pattern-specific error in a highly constrained physical model.

Portland, Oregon Area

Member of team that prototyped and developed the DL network used by the Calibre Machine Learning Optical Proximity (MLOPC) toolkit, that reduced time to train by 75% while maintaining optimal solution accuracy.Authored specs, testing plans, and implemented scripts to evaluate and authenticate the performance of the new HPC compute platform.

Hillsboro, Oregon, United States

Implementation and development of C++ - based algorithms and metrology solutions (recipes) to enable advanced optical photolithography fabrication techniques.Use of statistical DOE, data analysis techniques, and presentation skills to engineer integrated software solutions for complex, highly constrained problems under schedule pressure required to enable Moore’s Law.Owned five critical HVM photomask fabrication software metrology modules that verified mask manufacturability, for five generations of Intel processors.Single point of contact responsible for aligning multidisciplinary stakeholders to photomask manufacturing risks caused by RET engineering errors.

Ph.D., Physics

Activities and Societies: NSF Graduate Research Fellowship-Developed a novel characterization technique for determining the size.

Bachelor, Physics

Bachelor, Anthropology

Activities and Societies: Sigma Xi-Conducted primate fieldwork in Africa, and archaeological field work at Cahokia Mounds.

Education record