Welcome

Moore’s law is dead, and the age of silicon technology is closing. Quantum computing, its would-be alternative technology, is far from commercialization.

We believe that the only way to solve this problem lies in hardware architecture, because we are seeing huge amount of inefficiency in the architecture of existing processors. Even though the core function of the computer is naturally “computation”, only a small fraction of overall hardware is used for arithmetic units in the processor. In addition, the utilization of the arithmetic units are typically extremely low, producing their outputs only once in several hundred clock cycles on average.

Suppose an architecture in which most of hardware resource is used for building arithmetic units and those units are fully utilized. Systems computing thousands times faster than processors could be designed using the same amount of transistors. It also means that extremely low power consumption system could be viable for the same computational performance as the conventional processors.

Neuron Machine architecture, the proprietary technology of Neurocoms Inc., is a design paradigm in which the hardware circuit is designed according to the computational model it is supposed to compute, instead of compiling computation model (program) to fit the fixed hardware circuit, as processors do. As a result, highly efficient neural system can be designed by using most of hardware resource for arithmetic units and utilizing the arithmetic units for more than 90% of their clock cycles. We are excited to open the new era of single-chip supercomputing, replacing GPUs and supercomputers with highly efficient single-chip solutions

Applications

Deep Runner

Neurocoms Inc. | |
---|---|

Small-sized low-power consumption deep learning | |

Largescale Neuromorphic system | |

Matrix computation | |

Fluid mechanics (grid computation) | |

Computation-intensive image processing |