Technology

Our main products are Advanced Reconfigurable Computing Chip and its technology. The chip can run the computing power of ASIC and the flexibility of GPU under the higher cost performance than the ASIC at the corresponding time period. Through the reconfigurable technology, the chip can be upgraded along with algorithm for the low cost; through the technology of computing and memory integration, it can largely improve the computing power and reduce the delay, providing a computing power platform for 5G and low-latency AI computing.

Our products can be widely applied in cloud computing, autonomous driving, smart security and other fields. In cloud computing, our technical team is optimizing product prototypes, combining the leading AI chip technology with the demands of industrial clients. At this moment, the products have completed prototype verification and are in the stage of verification & optimization for small batch size. 

TensorChip provides integrated technical services of chip and computing and assists clients to incorporate their own algorithms with applicable platforms for implementation and mass production. In the realms of FPGA-based and RISCV platforms, TensorChip has now supported clients to implement algorithms in the areas of industrial intelligence and assisted driving, besides, is currently serving more clients in achieving the integration of algorithms and hardware platforms.

Tensorchip has established cooperations with following companies in RISV-V, industrial intelligence and smart assisted driving fields.

•   In the RISC-V field: such as Giga Device, Nuclei System Technology and Ali T-head Semiconductor.
•   In the AI field: such as Zhaoxin and Siasun Robotics. 
•   In the smart assisted driving field: we work with driving technology companies based in Beijing and Shanghai.

Reconfigurable Memory-Computing

The core of TensorChip: Evolutionary Reconfigurable Memory-Computing.

•   Its architecture surpasses the third-generation TPU, with its computing power and flexibility significantly functioning better than the normal AI chips (ASIC).

•   Under the circumstance of higher cost performance than ASIC chips at the corresponding time period, it supports the most advanced algorithms. (Figure 1)

Figure 1. Reconfigurable Memory Computing

RISC-V / AMR AI Acceleration

TensorChip's RISC-V /ARM AI acceleration technology can increase AP's AI computing power by about 25 times. (Figure 4, the value depends on the specific algorithm and AP float). If further combined with TensorChip's compression acceleration technology, the AI ​​computing power can be increased by about 50-100 times. (The value depends on the specific algorithm and AP float)

Figure 3. Supportable acceleration type