TAIPEI -- When Albert Liu convinced a friend to quit his high-paying job at Samsung Electronics in early 2015 and start a company together, the veteran Taiwanese chip developer did not realize that the plan would quickly go awry.
"An older friend had told me that there was already a promising business plan and financial sources lined up, and we just needed to execute that plan," Liu, the founding chairman and CEO of AI chip startup Kneron, told Nikkei Asia. "But those promises never materialized."
That left Liu, who was living in San Diego and working for Qualcomm at the time, in a difficult situation. "My friend had already quit his job, flown over, and was living in my house. He pushed me every day to take responsibility. ... I even later paid him half of my Qualcomm salary."
With no plan and no funding, Liu and Hao Kangli -- now Kneron's chief of software engineering -- had to come up with their own business idea. The pair spent long days in Liu's apartment or at a local Buddhist foundation, "discussing and programming things while other people were chanting," Hao recalls.
They eventually set an ambitious goal: to develop chips that would give devices like smartphones and self-driving cars AI capabilities.
Liu's vision proved prescient: AI-powered consumer electronics are now one of the hottest trends in global tech. Apple and Huawei kickstarted the trend by putting AI functions on their phones in late 2016. Huawei even approached Kneron in its early days and offered to buy its AI algorithm source codes -- an offer Liu declined.
Today, tech giants like Google, Apple, Intel, Nvidia, Qualcomm and Amazon are all developing their own edge-computing devices and services with AI capacities. Global revenue for AI chipsets used in edge-computing devices will grow from $7.7 billion in 2019 to $51.9 billion by 2025, according to forecasts by U.K. research agency Omdia.
Kneron on Tuesday secured strategic investment from Foxconn, the world's biggest electronics contract manufacturer and a key Apple supplier, and Taiwanese memory chipmaker Winbond. The amount was undisclosed. Winbond's subsidiary Nuvoton Technology, which just completed a deal to buy Panasonic's chipmaking unit, makes microcontrollers, controllers and audio- and power-related chips. Kneron said the investment tie-up will help the startup to collaborate with Foxconn and Winbond in the development of industrial and automotive applications.
Edge-computing, as the concept is called, was relatively new back in 2015. At the time few people recognized the potential, or even the possibility, of putting AI processing capabilities onto devices themselves. The assumption was that features such as speech and facial recognition needed to be carried out by powerful data center servers, which devices accessed via cloud computing.
"When I was proposing the idea in 2015 and 2016, no investors or clients really bought it. Maybe it wasn't the right timing. It was very challenging," Liu said. "I also didn't have the same support as some of our Chinese competitors, who could secure sky-high valuations and receive great government support. Investors in Taiwan are quite conservative compared with China, even if Taiwan has a huge chip ecosystem."
Kneron is, perhaps surprisingly, one of the only globally recognized Taiwanese chip startups. The island boasts the world's second-largest semiconductor industry by revenue, but the local ecosystem is dominated by giants established two and three decades ago, including Taiwan Semiconductor Manufacturing Co., MediaTek and United Microelectronics Corp. It is a stark contrast to China, where generous government support has allowed chip and AI startups to flourish -- the country already has two AI software unicorns, SenseTime and Megvii Technology, as well as its own AI chip national champion, Cambricon Technologies.
Nevertheless, when Liu set up Kneron in San Diego in late 2015, he decided to make Taiwan its research and development base. Bringing in talent and money was not easy, and he had to work hard just to bring about 35 people on board in the first year. Liu even went on a Chinese reality TV show called "I am an entrepreneur" to promote his startup. The constant international travel, however, meant less focus on management, and Kneron was in dire straits by the end of 2016.
"That was the darkest moment of my life," Liu said. "We were burning through cash and the company nearly shut down. When I returned to the office from my trip to China, only two people were still coming into work. All the others had angrily resigned."
Liu said he still remembers clearly how he spent a week calling up essential members and begging them to come back.
Hao, for his part, began to question whether the company had a future.
"I once also doubted whether this company could survive, but I think an entrepreneur will never give up and I saw that characteristic in Albert. I thought, 'I can trust this guy.'"
With Kneron's survival hanging in the balance, Liu decided to license its intellectual property for chips to Chinese home appliance maker Gree, its very first customer.
Licensing was a last resort, but before long, AI tech development began to flourish globally, and Kneron was able to secure several rounds of funding. Between 2017 and 2020 it raised more than $73 million at a valuation of up to $500 million, according to data provider Crunchbase, making it the most valuable Taiwanese chip startup. It already counted Horizons Ventures, Qualcomm, Sequoia Capital and SparkLabs Taipei as investors before its latest fundraising round.
That financial support allowed Kneron to put its own chips into mass production. Its KL 520 and KL720 chip solutions, unveiled in 2019 and in 2020, have been well-received, with international research agencies, including EETimes and Gartner, putting them on a level with Intel's and Nvidia's offerings. Kneron's facial recognition algorithm, moreover, earned the highest score in a prestigious competition held by the National Institute of Standards and Technology of the U.S. in 2019.
Kneron, whose workforce has grown to 160, has also managed to attract several high-profile chip industry veterans over the past five years. Former Qualcomm vice president of business development Adrian Ong is now the startup's chief commercial officer, and Lee HsiangTsun, a former Qualcomm director specializing in multimedia and computer vision, is its chief scientist. Davis Chen, former Qualcomm senior director of engineering, joined Kneron in early 2020 as vice president of engineering, while former senior sales director at Nvidia Alex Lo has also joined the chip startup.
Kneron currently has more than 30 customers, including the biggest European smart devices maker Sigfox, computer maker Asus, and internet companies such as Tencent and Sogou. Its chips are used in facial recognition door locks, computers and smartphones, as well as smart speakers, smart meters, connected vehicles and security cameras.
Edgar Chiu, co-founder and managing partner of SparkLabs Taipei, a venture capital fund and startup accelerator invested in Kneron, said the startup is setting an important precedent for the island. "It's great that finally Taiwan has more hardcore startups with real technologies and they are willing to go global," Chiu said. "It's been a long time since there were any hardware and chip-related startups for Taiwan, a place famous for its hardware and tech supply chain ecosystem."
According to Liu, U.S.-China tech tensions offer an opportunity for even more overseas growth.
"We see great growth opportunities in the global surveillance market as many of the players out there are looking for non-China options due to the geopolitical tension and the campaign to reduce [reliance on] Chinese technologies," he said. Chinese suppliers such as Huawei Technologies and Hikvision previously dominated the market.
In a recent briefing, Kneron said its revenue "could grow fivefold for 2020, and a further eight times in 2021." But for Liu, success is about more than his company's bottom line.
"I hope we are not only starting up a company but really starting up a new industry for Taiwan, a new AI industry, as there have been no new industries evolving here for a very long time," he said.