ArchiTek Launches AiOnIc®, the Industry's Most Compact, High-Performing Edge AI Processor
AiOnIc combines a 12 nm fabrication process with ArchiTek Intelligence®, a proprietary image processing architecture, to deliver a de facto standard SoC for secure edge-to-edge computing
MENLO PARK, Calif., March 4, 2021 (Newswire.com) - ArchiTek Corporation today announces "AiOnIc", its first in-house created chip based on the proprietary ArchiTek Intelligence® Pixel Engine (aIPE) architecture. The ArchiTek Intelligence architecture is at the forefront of Edge AI processors currently in development around the world. In a NEDO funded project[1] using SLAM, an application essential for self-driving vehicles,[2] AiOnIc reduces processing time to 1/20th of the speed of general purpose CPUs. Similarly, when performing OpenPose, an algorithm that calculates skeleton and pose estimation, AiOnIc achieves operating speeds approximately four times faster compared to that of the world's leading GPUs.
Realizing an ultra-compact, low power and high efficiency "fanless" processor
AiOnIc is the first prototype LSI using ArchiTek Intelligence, proving this high performance architecture can be packaged in a low-power, compact 12 nm silicon. AiOnIc does not need a cooling fan, so a sealed water-proof and dust-proof system can be easily designed and built.
ArchiTek Intelligence supports a variety of algorithms by dynamically recombining hardware components of CPUs, GPUs and application specific LSIs, and then enables the simultaneous execution of multiple processing in real time and with low latency. Due to its highly efficient hardware processing, ArchiTek Intelligence is ideal for use in integrated systems, such as smart security cameras, drones, robots, autonomous vehicles and other applications that have high-performance and low-power requirements for Edge AI.
AiOnIc Specifications
・Process: 12 nm (N12FFC) TSMC
・Package: FCCSP, 12 mm × 12 mm
・Die size: 4.5 mm × 4.5 mm
・Supply voltage: 0.9 V, 1.8 V
・Power consumption: Under 1W
・Built-in CPU SiFive RISC-V[3]
・Built-in SRAM 8MB
・Computer Vision Acceleration
・General purpose matrix adder and multiplier
・High efficiency General Purpose GPU (for parallel processing)
"AiOnIc delivers the holy grail for Edge AI processors by delivering true high performance with extremely low power at a fraction of the price of GPUs," says Shuichi Takada, ArchiTek CEO. "Our groundbreaking ArchiTek Intelligence computer architecture, combined with our low-cost AiOnIc processor, opens the door to deliver the promise of AR, VR, smart cities and countless other 21st century use cases by proliferating AI on the Edge."
For more information and availability inquiry, please contact Hassan Toorabally, CIO Global Business at [email protected] or visit www.architek.ai.
About ArchiTek
ArchiTek Corporation is a start-up venture engaged in the research and development of simple and beautiful architecture and algorithms. It is developing Edge AI processors that will form the nucleus of digital transformation in the areas of smart cities, smart factories, smart retail and smart agriculture. Formed in 2011, ArchiTek aims to expand from Japan into the global market as a platform for AI technology. For more details, please see our website at www.architek.ai.
Legal
ArchiTek, the ArchiTek logo, ArchiTek Intelligence and AiOnIc are registered trademarks of ArchiTek Corporation.
###
[1] These results from research commissioned by New Energy and Industrial Technology Development Organisation (NEDO), a Japanese government agency: www.nedo.go.jp/news/press/AA5_101318.html
[2] Simultaneous Localization and Mapping (SLAM) is a general term for technologies that simultaneously estimate the self-position of a moving object and create an environmental map.
[3] "ArchiTek select SiFive and DTS-Insight To Enable Next-Generation AI Solution Development" press release: www.sifive.com/press/architek-select-sifive-and-dts-insight-to-enable-next-generation
Source: ArchiTek
Share:
Tags: AI, AI Chips, Computer Vision, Computing, Edge AI, Embedded Computing, GPU, Heterogenous Computing, LSI, Microprocessors, Semiconductors