MN-Core - A Highly Efficient and Scalable Approach to Deep Learning
Ken Namura, Johannes Maximilian Kühn, Tamotsu Adachi, Hiroto Imachi, H. Kaneko, Tetsu Kato, G. Watanabe, Natsuko Tanaka, S. Kashihara, H. Miyashita, Y. Tomonaga, Ryosuke Okuta, Takuya Akiba, Brian Vogel, S. Kitajo, F. Osawa, Kuniyuki Takahashi, Yuki Takatsukasa, K. Mizumaru, T. Yamauchi · 2021
MN-Core is a highly efficient deep learning training accelerator reaching in excess of 1 TFLOPS/W (half-precision) at board level in real-world mixed-precision workloads. To reach and sustain this level of performance, the design is partitioned and packaged as four-die MCM package exceeding 3000mm2of die area.