Deep Learning Lead - DSP

Job Description:TI is considered to be one of the technology leaders in DSPs and purpose designed accelerator for high performance, low power embedded systems. These technologies part of our processors product portfolio fuels innovative and differentiated solutions in area of Machine learning and AI. TI Processors has a strong and robust software development team who focuses on solving some of the most challenging technical problems in automotive and industrial systems. TI EdgeAI technology offers a practical embedded inference solution for next-generation vehicles, smart cameras, edge AI boxes, and autonomous machines and robots. If your passion lies in creating optimized embedded systems with software innovations in cutting edge AI technology then, TI is the right place for you!As a Software Tech Lead - You will find yourself in middle of solving challenging problems in embedded systems. You will be working with small size, high impactful team who is responsible to deliver deep learning solutions on TI processors. Your work will involve design and development of deep learning inference software targeted towards ADAS and automated driving systems, robotics and industrial analytics. Your role demands you to work across the lifecycle of a software products from requirement gathering, planning, design, coding and testingBasic qualifications:Master's degree (or equivalent experience) or PhD in Computer Science, Electronics, Electrical Engineering, Artificial Intelligence, or Applied Math 8-12 years of industry experience in embedded system and software in domain of machine learningSkills needed:Great programmer: Programming skills in C/C++ and python, Modular and Object Oriented programming skills, familiarity with build systems - make, cmake, familiarity with LinuxIn-depth knowledge of embedded systems - VLIW and SIMD processor architecture, DMA, cache, memory architecture, inter process communicationWorking experience in machine learning technologies such as CNN, transformers, quantization algorithms and approaches for camera-based applications on embedded systemsWorking experience with DSPs (preferably TI DSPs) and hardware development boards/EVM for image/vision-based processing algorithmsGood knowledge on machine learning frameworks (PyTorch), inference solution and exchange formats (ONNX, ONNX RunTime, protobufs)Basic knowledge of RTOS and Linux with exposure to debugging of embedded systems - familiarity with heterogeneous core architecture is added advantageWell verse with software development life cycle and efficient use of associated tools - Git, JIRA, bitbucket, Jenkins, containers (Dockers), CI/CDStrong Communication, documentation and writing skillsSince Artificial intelligence and machine learning is very vast field and it requires varying expertise in different domains such as data scientist, cloud infrastructure an dev ops, model design and training, model deployment on embedded device or cloud and many more. Each of them itself is an area to build career and expertise. However, this role is more focused on deployment of models on embedded devices. Expertise on optimizing NN layers/ operators for embedded devices such as DSP or NN acceleratorsGood insight about critical instruction set to optimize ML model inferenceExpertise on data flow and memory management for optimal execution of ML model on an embedded SOC with DSP/accelerator and connectivity to on chip memory and external memory.Experience with utilizing on chip memory best and minimize access to external memoryWorking experience with DMA, good knowledge on processer architecture such as cache, DMA, memory banking structure etcExpertise in scheduling ML model across multiple cores and understanding data dependency challenges to achieve itGood knowledge on quantization of ML models for fixed point inference and associated challengesGood programming skills with C/C++ and PythonPrimary responsibilities:Design and Development of deep learning inference software on TI DSPs and accelerators. Analyze performance, identify optimization areas and optimize the software to entitle the hardware capabilities. Integration of this optimized inference solution with opensource run time frameworks such as ONNX Runtime /TFLite RuntimeDelivering software that scales across multiple SOCs: different OS, single to multi core architectures, different memory architectureAnalyzes and writes test and procedures for the software developedManage software deliverables from end user point of view - ease of use, well documented code, design documents. Participate in customer discussion and understand their requirements and implement it in softwareDevelop application specific software demos and examples related to end equipment which enables to benchmark TI processor capability and help customer to get started on their product developmentWork with hardware architecture and design teams to feed improvement areas for next generation DSPs and accelerator for machine learning applications. Also work on pre-silicon software development environments such as simulators, Zebu PRB