MILPITAS, Calif.—As AI is increasingly integrated into applications at the edge, it becomes more important for AI to conserve energy on devices by combining high reliability and precision with the efficiency of low power use.
Gyrfalcon Technology Inc. (GTI), a developer of high performance AI accelerators, is aiming to do just that with its new AI Accelerator Chip, the Lightspeeur® 5801. The chip boosts edge AI performance while using low energy, according to a release from Gyrfalcon.
The Lightspeeur 5801is the fourth chip in the company’s Lightspeeur portfolio of chips. All four share the Matrix Processing Engine, reported to provide the “best ratio of high performance AI with low energy use in the semiconductor industry.”
Offering 2.8 TOPS of performance while using only 224 mW of power, the chip provides what is reported to be the highest ratio of performance to power use for edge AI—12.6 TOPS/W. The Lightspeeur 5801 has four times the image input size when compared to the Lightspeeur 2801, the first chip introduced by Gyrfalcon in 2017. It allows higher resolution images to be supported and extends the range of accurate input capture for analyzing captured sensor data at the edge, the company said.
Clock speed has also been enhanced to support 50-200 MHz, which gives application developers a greater range of performance for their edge AI solutions. With less than 4 milliseconds of latency, the chip is said to deliver results with “blazing” speed, an advantage of edge AI processing over cloud AI processing.
Applications targeted by customers already using Lightspeeur chips are reported to include image recognition; object detection and tracking; natural language processing and understanding; and business intelligence. They also include facial recognition and visual analysis for consumer electronics; smart home and office; smart city; and industrial, enterprise, and data center applications.
In addition to the performance and power advantages, the Lightspeeur 5801 has a small chip size of 6mm x 6mm. It also has a built-in USB interface that can eliminate the need for some interface components on device designs, the company said.
The chip, intended for mass market edge AI device designs, is designed to reach even lower costs with high volume applications. According to Gyrfalcon, customers can get high performance AI for the premium applications in their device designs without opting for more expensive processors with built-in AI blocks.
“Customers can now integrate premium edge AI, without paying for the premium System-on-a-Chip on their edge device,” said Bin Lei, senior vice president of marketing and sales for Gyrfalcon Technology, Inc., in the release. “Our customers are proving to themselves that they can opt for mid to low tier host processors, combined with one of our inexpensive chips, to get the performance they need while saving $20 to $30 on their BOM costs.”
LG, which launched the mid-tier Q70 smartphone in October, is said to use the Lightspeeur 5801 to deliver premium imaging and video features that are otherwise only available on the most recently introduced premium smartphones.
The Lightspeeur 5801 is offered in a development kit called the 5801 Plai+ Plug, available on GTI’s DevPortal (https://dev.gyrfalcontech.ai/register/). The DevPortal also provides access to tools to build and deploy AI.
AI models can be trained with a model development kit. Applications can be developed with a Software Development Kit (SDK), which is available for Linux X86_64, Microsoft Windows, Android platforms, and ARM v7l and ARM v8 instruction sets.
Gyrfalcon Technology (www.gyrfalcontech.ai/) is a developer of high performance AI accelerators that use low power and are packaged in low-cost and small sized chips. The company received a 2019 Best Practices Award from Frost & Sullivan for its AI accelerator chips.