Page 26 - EETimes Europe June 2021

26 EE|Times EUROPE

Synaptics Makes a Natural Progression to Edge AI

Systems LLC and Marvell Technology Group’s ing and new video interface products and new seen in this space who have some sort of
Multimedia Business Unit. Conexant gave wireless solutions. novel concept about some kind of AI process-
the company access to advanced voice- and ing and are developing a chip that they want
audio-processing solutions for the smart BUILT ON EDGE PROCESSING to go broadly across multiple markets.”
home, including far-field voice technology for Processing at the edge is not new to
trigger-word detection and keyword spot- Synaptics. All processing of the raw data in NEXT STEPS
ting, while Marvell’s Multimedia Business its embedded sensing chips, including finger- In the early days, Synaptics developed its own
Unit delivered extensive IP for advanced print products and touch controllers, happens algorithms for its chips. The keyword-spotting
processing technology for video and audio on-chip because of concerns around power, and trigger-word algorithms for voice digital
applications, particularly digital personal latency, and security. assistants, as examples, are the company’s
assistants, as well as the smart home. Even before Synaptics manufactured its core algorithms. But Synaptics wanted to
With the Conexant acquisition, first interface products, the company was open up its silicon to allow third parties to
Synaptics gained a portfolio of audio prod- founded in 1986 to research neural networks run their own deep neural networks and other
ucts, providing the right architecture to do but pivoted into other technologies before algorithms on its chips, so it needed a tool
keyword spotting at the edge — and that is coming full circle to develop edge-AI suite. That is not so easy to do.
done through neural networks, said Worfolk. processors for computer-vision and multime- The company entered into a partnership
The multimedia team carved out from Marvell dia applications. with Eta Compute to develop the software
Technology was developing video proces- “We were founded to do neural network tools to train a deep neural network and
sors, and those devices are used in streaming chips over 30 years ago; back then, all the compile it “so it can run on our silicon, and
media as well as in smart displays, he added. chips were analog AI chips, and it was chal- we could move a little faster and open up our
“As these smart display products integrate lenging to scale well,” said Worfolk. “In fact, chips to third parties,” said Worfolk.
cameras, you run into all the same challenges the company went off in a slightly different There are other challenges in a market
around performance and privacy, and there’s direction after the initial founding and started where innovation is happening at such a fast
more and more drive to do those algorithms doing pattern recognition, which is a classic pace that performance tradeoffs often result.
in the edge device,” said Worfolk. “The natural AI problem. We’ve been doing AI for a long “The field as a whole is very immature, and
structures for those types of algorithms today time, but we have recently migrated to deep in that sense, it is moving very quickly; there
with the best performance are AI algorithms.” learning, and these deep neural networks are new announcements about new types of
In 2020, Synaptics bolstered its IoT position have really taken over by storm. neural networks every single week, and a lot
with the acquisition of Broadcom’s wireless “With the breakthroughs in AI in the last of the work has been done through academic
IoT business, adding Wi-Fi, Bluetooth, and decade, more and more of these traditional or big research groups that are trying to
GNSS/GPS technologies for applications algorithms have mapped over to AI algorithms, push the boundaries of performance,” said
including home automation, smart displays enabling performance advantages when you do Worfolk. “But there is a big gap between
and speakers, media streamers, IP cameras, the processing at the edge,” he added. academic research and what can actually run
and automotive. By pairing its edge SoCs with Worfolk said the nature of the company’s on a small device.
the wireless technology, it can open up oppor- products and the vertical markets it serves are “Although we are seeing algorithms that are
tunities beyond the consumer IoT market. driving the need for AI-based algorithms. able to perform on new levels that we’ve never
Synaptics also acquired DisplayLink Corp., “We’ve entered the AI space vertically seen before on the vision or the audio side,
adding that company’s universal docking through our existing markets, and then they take more and more compute.”
solutions and video-compression technology with those products, we’re expanding into This often translates into tradeoffs between
to its portfolio. Synaptics will combine the neighboring markets,” he said. “This is quite efficiency and flexibility. What typically
video-compression technology with its exist- different from many of the startups you’ve happens is that the first piece of silicon for
a particular target market isn’t very flexible,
and as the neural networks that “run on that
silicon mature and become more stable to
produce the desired functionality, we can look
at a second-generation chip,” which is opti-
mized for that performance, said Worfolk.
“It’s what makes it so exciting for us,
because there’s always something new and
interesting, but it is a challenge from a busi-
ness perspective,” he said.
The company’s strategy is to build the sil-
icon and then create an example application
or demo that makes it easier to discuss with
customers. “It also generates ideas around
what you can do with the chip and makes sure
that we understand what it takes to bring this
chip to product,” said Worfolk.
At the Embedded Vision Summit last month,
Worfolk demonstrated the Katana chip in a
battery-powered people counter used to track
usage in office conference rooms.
At the Embedded Vision Summit, Synaptics demonstrated the Katana Edge AI SoC in a “This system is not just the Synaptics
people-counting application like this one. (Source: Synaptics) Katana chip; it also includes cameras, motion

JUNE 2021 | www.eetimes.eu