In mid-November of 2016, on the eve of the Thanksgiving holiday, Albert Liu, the 35-year-old founder of AI chipset startup Kneron, jumped on a flight and rushed back from China to the company’s San Diego headquarters, hoping to get back in time to thwart a mass resignation by company staff.

When he got there, only two of around a dozen U.S. office staff had come to work that day. Among them was Kidd Su, a former Broadcom engineer and friend of Liu’s from college. Su had only been at Kneron for four months when the prospect of a mass resignation of his colleagues emerged.

That year, 31 of 36 staff members across Kneron offices in the U.S., Taiwan and China resigned.

“Everyone was feeling that the company lacked direction,” Su recalls. Liu was often flying around between China and the U.S. looking for customers, and “...didn’t pay much attention to running the company. Sometimes he’d return, only to change the direction of product R&D at the last minute because of a client’s idea. The team started asking ourselves: What are we doing here?”

CB Insight: Kneron Among World’s Top 100 Most Promising AI Startups

Founded in 2015, Kneron nearly fell apart the next year. Liu’s stress was so great  that he was plagued with insomnia, and even threw up on the street.

But four years later, in January of 2020, Kneron obtained a US$40 million round of funding from Li Ka Shing’s Horizon Ventures, raising its total A Round funding to $77 million.

Kneron’s AI system on a chip (SOC) has already been adopted by the likes of  Broadcom, Elan Microelectronics, AAEON Technology and co-developed an AIoT solution with French Internet of Things (IoT) leader Sigfox.

This March, Kneron was added to CB Insight’s “AI 100,” a list of the 100 most promising startups in the artificial intelligence field.

Enduring the tumultuous storm of 2016 was critical to getting to where Kneron is today.

“Some colleagues thought I wanted to be a celebrity,” recalls Liu.

Liu appeared on a Chinese reality TV show called “I, Founder” in 2016, believing it would help the company’s marketing. However, this gave a number of company staff the idea that he was not prioritizing company affairs. Combined with the lack of expected upswing in orders, it contributed to a flood of resignations instead.

Panicked, Liu sought counsel from fellow show guests Dong Mingzhu, chairwoman of Gree Electric, and Wang Xiaochuan, founder of China’s number two search engine, Sogou, admitting to them that “the company is nearly bankrupt.”

Dong and Wang responded with around US$2 million in orders over the next two years, providing the company with a measure of timely relief.

Liu also reached out to former MStar Semiconductor co-founder Steve Yang for help. Yang’s then-current company, the venture capital firm CID Group, provided some liquidity in the form of a US$3.5 million convertible loan. The capital in place enabled Liu to build a new team.

Orders and capital provide sustenance, and Liu was able to retain around nine staff members while continuing to expand recruiting. Most notable among the new recruits was Hsiang-Tsun Li, a 17-year veteran of Broadcom.

Currently serving as Kneron’s chief scientist, Li and Liu go back to Broadcom days in the United States, when Li was Liu’s “boss’s boss” as Broadcom’s head of multimedia R&D. Speaking to CommonWealth, Li said he saw great potential in Kneron’s formidable AI software development team.

Kneron co-founder Mau-Chung Frank Chang, Liu’s adviser at UCLA and former president of National Chiao Tung University, credits Kneron’s strong technical foundation for getting the company through its 2016 crisis. As a result, having reached the brink of collapse, the company was able to quickly regroup with an infusion of capital and manpower.

While teaching at UCLA, Chang was impressed with Liu’s technical acumen, describing Liu as “one of the few among the 50 or so Ph.D students I advised while teaching in the U.S. who possessed the ability to define both products and markets.”

Elaborating, Chang related that the U.S. government and corporations often look to schools for graduate students to help solve problems, “but Liu chose to work on his own content and wanted to develop products, not just help others solve existing issues.”

Prior to pursuing his doctorate at UCLA, Liu had worked as an engineer in Broadcom’s media R&D center, where he was involved in deep learning, autonomous driving systems, and 3D facial recognition operations.

Chang credits this experience for developing Liu’s market sensitivity in addition to his R&D skills.

Accordingly, Liu decided to take on a major problem for his Ph.D thesis work: how to save energy consumption when AI in assorted IoT devices is performing facial and image recognition operations.

The result became Kneron’s main product: low power consumption AI systems on a chip (SOC).

Kneron introduced its first-generation chip product last May, the KL520, equipped with a dual core ARM Cortex M4 CPU and autonomously developed NPU to go head-to-head with Intel’s Movidius visual processing unit (VPU).

Kneron has calculated that for the same-size image recognition, its computing power is 0.34 TOPS, and, despite being only half as fast as the same class Movidius chip, the Kneron is more energy efficient, requiring only one-third the power compared to Intel.

Kevin Chiu, vice president of Kneron customer AAEON Technology, reveals that even though AAEON also offers the option of Movidius edge computing modules, thanks to the energy-efficient, high price/performance ratio advantages Kneron boasts, “(they can) reach markets that Intel cannot access; some (end) customers specify Kneron by name,” he says.

For instance, one French customer in the engineering field wanted backhoes to be able to automatically recognize obstacles and incorporate basic autonomous-driving functions.

Upon comparing specifications, they were immediately impressed by Kneron’s low power consumption, found recognition speed acceptable, and the cost/performance ratio high, and they asked Kneron to send samples.

The first impression Liu invariably gives is that of a “nerdy engineer”. However, years of paying visits to clients and talent show appearances have honed his speaking skills and even won him a group of supporters among his elders. These include former National Tsing Hua University president and Academia Sinica member C.L. Liu, who openly endorses Kneron.

Introducing himself before a speech at a forum held at Tsing Hua University in late July, Albert Liu emphasized “strong investors”. In addition to Sequoia Capital, Alibaba, and Li Ka Shing’s VC fund, before a full house of NTHU students and faculty, including Macronix International CEO Miin Chyou Wu, he let it slip that “Morris Chang’s wife, Sophie Chang, is also one of our investors.”

Kneron introduced its second-generation chip, the KL720, earlier this year, packing three times more computing power and 0.9 TOPS/W to keep competing with the latest generation chip brought out by Intel this year, the Movidius Myriad X.

Many customers currently tout Kneron’s cost- and power-savings advantages. However, if industry giants like Intel decide to slash prices, or target small market segments like image controllers via internal task forces, Kneron would have a hard time remaining the top choice. 

One venture capital firm director, whose company is a Kneron investor, adds that in the future, good business models must go beyond “single technology competition”. For example, “Intel might not have the best CPUs, but it’s undoubtedly the best integrated company with Microsoft, which established a sound business model. 

Business model competition is not just a war of chips, but one of ecosystems: You have to bundle algorithms, data structures, IC design, and the application end together for sale in order to establish a competitive fortress,” the director said.

Going forward, finding niche markets and a sound business model is sure to be a major challenge for Albert Liu.

---

Have you read?
♦ Looking for Eco-friendly PET Bottles? Call Taiwan
♦ Server Chassis Giant Chenbro Keeps Talent in Taiwan
♦ Taiwan’s Top 50 Performers: IC Design Leads the Pack Thanks to China’s 5G Potential

Translated by David Thoman
Edited by TC Lin
Uploaded by Penny Chiang