Moffett AI's core technology, sparse computing, is globally recognized as a new generation of AI computing technology. Its principle is to eliminate zero-containing or invalid elements from a large number of matrix operations in AI computing, so that the neural network model can eliminate redundant and repetitive content, reduce computing pressure, accelerate computation speed, and improve effective performance. In the context of Moore's Law failing and hardware-supported arithmetic power increase approaching its limit, sparse computing provides a more ideal solution from the algorithmic level, which can effectively break through the current arithmetic power bottleneck.
Moffett AI is one of the earliest companies in the world to lay out sparse computing, and is the first in the industry to realize the sale of commercial sparse computing products. Moffett AI's technological advantages include two main aspects: the world's original double sparse algorithm and the synergistic design of software and hardware.
On the one hand, Moffett AI's technical strength ranks among the world's leading companies: Moffett AI is the inventor of the double sparse algorithm and the first company in the world to apply the double sparse algorithm; it owns more than 30 patents around the world. Moffett AI's core self-developed double sparse algorithm is able to realize 32-fold sparsification of AI models.
On the other hand, Moffett AI takes the lead in elevating sparsification from algorithm to the level of software-hardware synergistic design, defining hardware by sparsification algorithm, realizing hardware architecture innovation for sparsification technology, and constructing the underlying algorithmic capability to continuously optimize sparsity computation at multiple levels, and the architecture ensures the programmability, highly expandability and fast iteration capability, so as to give full play to the effect of sparsity computation, and enable the products to obtain groundbreaking performance Improvement.