Green and sustainable Technology: There is a sustainability angle to this innovation from Qpisemi. There is study in 2019 that points, that training a certain deep neural network for NLP (natural-language processing) produced five times the CO2 emissions typically associated with driving an automobile over its lifetime. Apart from this, intelligence modeling using neurons has large limitations. Addressing these two issues with AI20P solutions stack can make it greener and most efficient AI technology in the world. With upcoming metaverse applications and increasing complexity of intelligence modelling, AI20P is a global need today. With advances in integrated photonics, AI20P technology stack will soon be with system integrators across applications providing us Green and sustainable AI Technology.
Silicon Photonics process research: Qpisemi has an objective to create innovations in process research and fabrication. Silicon photonics-based process research will immediately yield results for its product such as AI20P. Its process research will help it in miniaturing many of optics component and integrate it with electronics efficiently. Qpisemi research in integrated photonics device is also aimed at improving reliability and hence shifting to lower process nodes.
Dr Nagendra Nagaraja revealed, “AI 2.0 is more advanced technology than current DL/ML technologies available. AI 2.0 would enable efficient actionable information generation in real time, that would match close to human intelligence at certain tasks at the edge. AI 2.0 would model emergent behavior accurately which is not possible current with DL and ML technologies. This would enable accurate modelling of megastructure like spread of pandemic, economies, transportation networks, advanced metaverse applications and advanced manufacturing. AI 2.0 will have immense effect on manufacturing, advanced metaverse applications and it would change the way we handle supply chain. Number of mathematical operations required in AI 2.0 and latency requirement of these operations cannot be met by current electronics-based ML processors and GPUs. We need something that is much faster. Matrix multiplications in photonics domain is 100x faster than in digital regime. Photonics domains offer massive parallelism that is required for AI 2.0. Photonics is the future of AI. Arithmetic operations in photonics domain using light almost consumes no energy and it is 100x faster, generate very less heat. Amount of energy consumed by current GPUs, ML training and inference chips are unsustainable, and we do require processors which consumes lower energy and provides large throughput that is required for AI 2.0. However, problem with silicon photonics chips is that they are large, integration with electronics is a challenge. We have innovation to address both of these issues. We have technology to scale photonics chip from data centers to edge.”
Dr Nagendra Nagaraja further said, “Although silicon photonics computational layer and design to miniaturize is major driving factor for AI 2.0 technology enablement, we have to design new kind of compilers, software libraries, newer application interfaces, also newer system engineering techniques to enable AI 2.0.”
Qpisemi has built strong team for silicon photonics and is currently hiring for hardware and compiler design. It is expected to have 50+ strong engineering and research by Q4 of 2022 and is expected to have first sample of AI20P available by end of 2022 or early 2023, sampling with key customers. In 2022 Qpi Technology (parent of Qpisemi) subsidiaries QpiAI and SuperQ are all scheduled to tapeout multiple chips and hardware to feed into market demand for quantum chips and quantum communication.
Pinakin Padalia, Co-founder and Director of Qpisemi suggested, “Along with discrete cryo electronics, which is getting taped out in 2022, we hope to have working chip for AI20P and also Quantum secure CPU ‘Prakhar’ at test chip and simulations levels respectively this year. We expect to build high quality R&D and business development teams at Qpisemi this year for scaling into production in 2023-25.”
