Kneron’s Latest KL720 AI Chip Aims to Expand Enterprise AI Use Cases
With its recently announced KL720 AI chip in early, low-level production, AI developer and chipmaker Kneron is busy imagining a wide range of customer use cases where the new silicon will offer promise in new markets.
The KL720 chips are designed to deliver lower power consumption, more processing power and increased security and privacy over competing chips from Intel, AMD and others, allowing them to power a myriad of new devices in the marketplace, according to the company. The KL720 chips arrived 15 months after the company’s first AI chips, the KL520, launched in May of 2019.
The KL720 SoC delivers 1.4 TOPS per watt and includes an Arm Cortex-M4 @400MHz for system control, a Cadence DSP @500MHz as an AI co-processor, and features average power consumption of 1.2 watts. The chip supports 4K images, Full HD (1080p) video, and natural language audio processing, enabling devices to capture more detail for facial and audio recognition. In addition, the KL720 can be reconfigured as needed and supports full natural language processing, giving it more capabilities for a wide range of use cases.
“In a lot of these AI chips out there, they are either very expensive or they are AI chips based on Arm architecture,” said Chen. “We are taking a microcontroller that uses much less power. Competitors are using A7, A9, A53 processors in smartphones and tablets. Others, including A72 Arm processors, all are higher profile and use a lot of power.”
A huge benefit of Kneron’s new chip is that while cutting power consumption needs, the chip provides a small form factor for use in a wide range of devices, said Chen. That means devices running the KL720 can include smaller batteries, which can help design and integration. This can be particularly helpful for devices deployed in remote locations, including power plants, wireless towers, smart meters, edge server applications and more, he said.
The new KL720 chips are just starting production now by the hundreds to thousands, with those chips going out to some of Kneron’s partners for evaluation and testing, he said. Devices using the new chips are expected by the end of 2021, according to Chen.
Karl Freund, senior analyst for machine learning and HPC for analyst firm Moor Insights & Strategy, said Kneron’s latest chip arrives as chipmakers are seeing markets for specialized AI chips that are smaller and more power efficient for a wide range of uses.
“OpenAI.org is saying that the size of the neural network models are doubling every 2.5 months,” said Freund. “Human brains have a few billion neurons. The rate of increase in neural network models is pretty shocking. It outstrips Moore’s Law by thousands of times.”
GPUs tend to be too expensive to run in smart city applications and similar uses, which has helped the AI chip market thrive with startups such as Kneron, he said.
“What you have is companies like this that see the opportunities,” said Freund. “This market will probably be worth $25B in the next five years. It will grow rapidly. And the chips you need for a self-driving car are much more complicated and expensive than the chips you put on every lamppost” in smart cities.
“It’s a whole new era of computing,” said Freund. “Two years ago, many of us predicted a Cambrian explosion in chip design.” What happened is that it exploded in PowerPoint presentations, not in actual chips at the time. Now the market is seeing these designs and new AI chips come to fruition, he said.
“It takes longer than they thought,” Freund said of AI chipmakers in general. “Now were start to see that change.”
Freund said he expects to see scores of new AI chipmakers on the market over the next few years, all competing on power, performance, and price.
“Think about the games your kids will be playing with in five years,” he said. “There will be a market for all of the chipmakers. It will absolutely explode. It will take more time than people expect, but then about 80% of those companies will drop off [and disappear]."
Based in San Diego, Kneron is backed by Alibaba, Qualcomm, Sequoia, and Horizons Ventures. The company was founded in 2015.