The CEO of AI chip platform startup Openedges Technology expects the company to experience double to triple growth per year going forward from high demand for its IP solutions.
The company launched its mixed-precision (4/8bit) computation neural network processing unit (NPU) IP dubbed Enlight last month.
Previous NPU IP for edge networks mostly used 8bit as lowering the bit could reduce power consumption and increase surface area efficiency could also lower accuracy, Openedges CEO Sean Lee explained.
Enlight, by processing as much computation in 4bit as possible while using 8bit for tasks requiring high-precision, can reduce power consumption while retaining the benefits of an 8bit NPU IP, Lee said.
Enlight has been verified in uses such as security cameras and advanced driver assistance systems, the CEO said.
The NPU IP can be used together with Openedges’ memory system IPs such as those for memory interconnector, DDR controller and DDR PHY, Lee said.
For NPUs to perform optimally, they must be paired with a memory system that can save and provide high-density data, the CEO said.
Openedges’ strength lies in the fact that it offers these various IPs in a suit, unlike competitors that offer IPs for NPU and memory systems individually, he said.
The startup aims to expand in the US, China and other major semiconductor markets.
The company was growing exponentially and it will stand shoulder-to-shoulder with top IP firms globally in the next four to five years, the CEO claimed.
Openedges was founded by Lee, who developed IPs related to DRAM at Samsung, and other chip experts in 2017. It offers semiconductor IPs, with its main product being memory systems and NPU IPs for edge networks.