Takeuchi Lab is studying Computation in memory (CiM), which combines data processing and memory like the human brain with a focus on data. Conventional computers require accurate calculations that do not allow errors. On the other hand, machine learning applications such as image and speech recognition allow some errors. By tolerating some errors, Approximate computing improves power efficiency and performance by 10-100 times. We are also working on high speed LSIs for AI processing (called AI chips), 3D LSI circuit design integrating LSI chips in 3-dimention, and datacenter scale computing which handles one database for the entire data center. With the motif of enhancement of memory reliability, “Explainable AI (XAI)” explains why AI can recognize image/speech by comparing results of AI and physical failure models. Aiming at the future of machine learning, we also started research on quantum annealing using the interference effect of nanodevices, brain-inspired memory that stores data flexibly like the human brain.
For more detail