With Korea's brightest chip minds, FADU eyes the next Broadcom

Home > Business > Industry

With Korea's brightest chip minds, FADU eyes the next Broadcom

Published: 31 Jan. 2025, 07:00 Updated: 31 Jan. 2025, 09:31

JIN EUN-SOO
[email protected]

Audio report: written by reporters, read by AI

FADU founder and CEO Lee Ji-hyo speaks during an interview with the Korea JoongAng Daily at the chipmaker's headquarters in southern Seoul in January. [PARK SANG-MOON]

[INTERVIEW]

As the smartphone market plateaued in the early 2010s, the global chip industry braced for a downturn, spurring an exodus of talented engineers who left hardware behind to pursue opportunities in software.

Software powerhouses like Google, Microsoft and Facebook eagerly absorbed this influx of talent, thriving on the migration.

In Korea, however, the shift was slower.

Samsung Electronics and SK hynix, offering some of the country’s most competitive salaries, managed to retain and attract top engineers, keeping Korea’s semiconductor sector stronger than ever.

This dynamic presented an opening for Lee Ji-hyo, founder and CEO of Korean chip design startup FADU.

"In Korea, the brightest professors and most skilled engineers were still researching semiconductors, even as the United States had largely abandoned them, Japan and Europe had stepped back and Taiwan was focused solely on manufacturing," Lee said in a recent interview with the Korea JoongAng Daily.

"Korea was the only country left with the potential to design next-generation semiconductors. Rather than entering established markets, we aimed to carve out a new one — SSD [solid-state drive] controllers for data centers."

Founded in 2015, FADU specializes in controller design for enterprise SSDs or eSSDs and is now venturing into cutting-edge technologies like CXL (Compute Express Link) and power semiconductors.

There are growing pains too. It is accused of violating the Capital Market Act when going public in 2023 in Korea's secondary market by allegedly inflating its expected earnings and concealing critical information from investors. Korea’s financial watchdog referred the company and a lead underwriter of the IPO to the prosecution in December 2024.

The company’s momentum, however, is undeniable: In 2024 alone, FADU secured 62.4 billion won ($50 million) in supply deals with major players like Western Digital and SK hynix. Its end clients reportedly include global giants such as Meta, Nvidia and SpaceX.

Although still unprofitable, FADU has demonstrated significant revenue growth, soaring to 10.1 billion won in the third quarter of 2024 from 321 million won a year earlier. Lee remains optimistic about this year, aiming for a confident turnaround in 2026.

The Korea JoongAng Daily sat down with Lee at the company’s headquarters in southern Seoul to learn more about his venture into the high-stakes world of eSSD controllers.

What is an eSSD controller?

In memory, there are two main types: DRAM [dynamic random-access memory] and NAND. DRAM doesn’t require a controller because it is inherently fast and reliable. The challenge lies with NAND. While NAND is cheap, it sacrifices functionality, making it prone to heat, slower performance and fragility.

With the rise of AI, however, the performance demands placed on NAND have increased, but the drive to make it cheaper every year has exacerbated its shortcomings. That’s where controllers come in. Think of NAND as a fragile water tank and the controller as a pump. The controller extracts water [data] from these tanks before they fail and ensures a stable flow to farmlands, which represent CPUs or GPUs. Controllers also manage the heat generated by NAND — if the water tanks get too hot, the controller prevents their use.

As NAND becomes cheaper, the role and functionality of controllers will become increasingly critical.

Why did you choose to focus on eSSD controllers back in 2015, when working on DRAM was mainstream in Korea?

I used to work as a partner at Bain & Company, leading the semiconductor sector. At the time, Korea wasn’t particularly strong in NAND production — there were six major NAND companies globally, and Samsung and SK hynix were ranked fifth and sixth.

I believed that data centers would continue to grow because computers would inevitably become smarter, approaching human intelligence. According to Moore’s Law, computational performance doubles every two years, and that point in time when computers exceed the intelligence of humans wasn’t far off. It was clear that this growth would require massive data processing, driving explosive demand for data centers. If the data center industry were to boom, eSSDs would play a key role, and at the core of eSSDs is the controller.

How did FADU manage to secure substantial deals with major players despite being a latecomer?

In 2015, the industry underwent a significant shift. Previously, data centers relied heavily on hard disk drives [HDDs] because they were perceived as more reliable. SSDs were mainly used in disposable devices like USB drives or smartphones, which followed a two-year replacement cycle.

The specifications for HDDs were relatively simple — like a thin water pump. But as clients began conceptualizing large-scale data centers, it became clear that these pumps couldn’t handle the vast amounts of data storage required. A new SSD-optimized specification called NVMe PCIe emerged, and after this specification became the norm, performance requirements for SSDs increased fivefold.

Legacy players couldn’t easily abandon their existing design methods, so they resorted to connecting five thin pumps together. From our perspective, this approach was inefficient — it generated excessive heat and lacked sustainability. So, we developed an entirely new architecture optimized for next-generation SSDs.

While the legacy players initially maintained their dominance, their designs began to falter as performance demands doubled every three years. Connecting five, then 10 and then 20 water pumps did not work anymore. When the industry transitioned to this new generation of requirements, we were the only company with a functional chip design.

Was the journey smooth after that?

This industry operates on a roughly four-year cycle although Nvidia is trying to break that pattern recently. For instance, chipmakers like Intel announce new CPUs every four years. This provides a road map, giving companies time to develop SSDs or other components tailored to those specifications.

About two years into the cycle, companies evaluate your progress — essentially asking, “Let’s see what you’ve done.” If you pass this checkpoint, they’ll ask you to produce a real product that can test their processors. Successfully completing this process allows you to supply your components stably for the next four years, after which things become more predictable.

In 2018, we manufactured our first chip, but no one wanted to use it. They didn’t know who we were, and while they acknowledged our technology, they weren’t willing to take the risk. It took three to four years testing the waters before we secured our first sales in 2022.

Since then, we’ve built substantial credibility with our clients, and we believe these relationships will endure over the long term.

How does the U.S. chip trade sanction on China impact your business?

The sanctions create both risks and opportunities. China’s chip market accounts for 20 to 30 percent of the global total. If we aim to be a major player rather than a niche one, we can’t ignore the Chinese market. While Chinese chips are still catching up to U.S. standards, this presents an opportunity for us to bridge the gap.

What’s your global strategy?

Our priority is to establish our technology in the U.S. market. Once we’re verified there, it becomes easier to penetrate other markets. We already targeted China by establishing a local operation there last year, and India is also on our radar. India has recently begun investing heavily in state-backed data centers to accommodate the AI boom.

How are your CXL and power semiconductor businesses progressing?

We expect tangible results from our power semiconductor business sooner than from CXL. We began quality testing with clients last year and plan to start mass production this year. For CXL, we anticipate needing two more years to bring it to market.

What’s your long-term vision?

Our role models are comprehensive chip design companies like Marvell Technology and Broadcom. To emulate them, we need expertise and products in the three main pillars of processors: memory, networking and computation. We’ve already secured our position in memory and networking, and we plan to venture into processors. Our ultimate goal is to become a comprehensive chip company specializing in data centers. Also, as a listed company, I am committed to fully embracing my responsibilities and meeting expectations by delivering cutting-edge technology.

BY JIN EUN-SOO [[email protected]]