Products Overview

-A-

Hard & Soft ASIC AI IP

The hard and soft ASIC AI IP will be available in foundaries IP library/catalogue

Plan to enroll into IP Alliance Program

FPGA System & IP

Deliver both hard and soft AI IPs for FPGA design.

Prototype and productise in FPGA

Edge Smart Devices

AI Development Board with pure ASIC AI chip

Edge AI systems and devices

Proprietary AI software engine

AIRIS Labs AI KNN (K-Nearest Neighbour) Design

The Artificial Intelligence (AI) field has been growing rapidly these past few years. In current systems, AI computing is mainly focused on cloud AI where many remote devices connect to the cloud for AI processing. However, the trend is swiftly evolving into edge computing on edge devices because of greater compute power and smaller size of the AI processor.

One of the key contribution of edge computing is the development of smaller size AI processors. In current market, many companies are focusing on utilizing the CPU, GPU, TPU & NPU to create an edge AI system. However, due to “von Neumann bottleneck”, those processors are typically not efficient in handling the AI task because each ALU can only execute one transaction at a time.

To overcome the bottleneck issue, many developers create AI chip that runs at high frequency and with many cores to allow multiple instructions or tasks running in parallel. However, this method will increase the power consumption and cost significantly.
To create a low power and high performance AI system, Airis Labs introduces the ASIC AI KNN (K-Nearest Neighbour) design, which can perform general purpose KNN algorithm in absent of software instructions or programming codes. As this is an ASIC AI design, it is free from “von Neumann bottleneck” issue.

Key features of Airis Labs AI KNN design:

High Performance

LIGHTNING FAST ~ Very fast KNN learning and classification. The design merely need ~0.7ms to perform classification when running with 100Mhz clock and on 1024 samples x 1024 attributes per sample, which giving 1 MByte of learning data.

LOW POWER ~ Power consumption is < 0.15W when running at 100MHz

No Coding

NO SOFTWARE ~ No special software or programming code needed to execute the KNN algorithm.

LOW POWER ~ Power consumption is < 0.15W when running at 100MHz

SILICON PROVEN

TSMC 40nmLP ~ Silicon proven on TSMC 40nmLP process utilizing full logic gates and SRAM memory.

VALIDATION & TESTING ~ Besides typical post-silicon validation, the team also built a AI shield, which connects to Raspberry Pi directly, to perform various type of classifications and proof the functionality of the design.