Before artificial intelligence can answer a question, an astonishing amount of information has to go somewhere else. Bits leave memory, cross a package, travel along a board, enter an accelerator, and come back carrying an answer. The arithmetic gets the applause. The journey gets Harry Chan.
Chan is the founder and CEO of eTopus Technology, a San Jose semiconductor-IP company devoted to the circuitry that makes those journeys possible. The field is called SerDes, short for serializer/deserializer. It takes a crowd of parallel signals, lines them up into a fast serial stream, sends them through a physical channel, and reconstructs them at the other end. It is part traffic engineer, part translator, part very stern customs officer.
That work has occupied roughly three decades of Chan's career. He studied engineering at the Hong Kong University of Science and Technology, completing a bachelor's degree in 1995 and a Master of Philosophy in 1997. By 2000 he was leading a SerDes team at LSI Logic. Later he became technical director at Link-A-Media Devices, the storage-chip company acquired by SK hynix. The logos changed. His obsession did not: moving more data, across less forgiving paths, without wasting power or time.
01 / The overlooked bottleneckThe road becomes the product
Computing used to reward the grand central monument: one larger chip doing more work. Modern systems are increasingly neighborhoods. CPUs, GPUs, memory and specialized accelerators divide the labor. Chiplets split complex designs into smaller pieces that can be combined in one package. Every split creates an interface, and every interface creates an opportunity for delay, noise, heat and error. The seams matter because the machine is becoming mostly seams.
eTopus works in those seams. Its public product material describes high-speed mixed-signal IP for Ethernet, Interlaken, PCI Express, CXL and die-to-die links, across applications that include networking, storage, 5G and AI. The company's architecture combines analog-to-digital conversion with digital signal processing. In plain English, it measures a battered incoming signal, applies mathematics to recover what was sent, and tries to do so before latency turns speed into theater.
A bit's commute
SerDes turns many lanes into a fast stream, carries it across an imperfect physical path, then rebuilds the original data.
Chan is careful about the phrase "real world." When eTopus worked with connector maker Hirose on a 112Gbps demonstration in 2021, he emphasized testing the IP through an actual connector and bench setup. That distinction is the hardware founder's tax. Code can be patched after launch. Silicon carries the decisions of months ago into a laboratory where physics gets the final vote.
His publication list reads like a compressed history of where the bits have needed help. Early work examined low-power motion estimation for video, where saving computation meant saving energy. Another paper covered a 700 Mb/s BiCMOS read-channel integrated circuit. Later work addressed interference in shingled magnetic recording, a storage technique that squeezes tracks close enough to overlap like roof tiles. The subjects look separate until the common problem comes into focus: information must be recovered from a physical world that smudges, weakens and crowds it.
This is why "connectivity" undersells the job. A wire does not guarantee understanding. Higher data rates make every imperfection louder. Channels lose energy. Adjacent signals interfere. Timing wanders. Receivers have to decide what the sender meant, often from evidence that looks more like weather than a neat sequence of zeroes and ones. Chan's career has followed that decision deeper into mixed-signal design, where analog reality meets digital correction.
“When developing high-speed IP solutions, it is important to test them in a real-world environment.”Harry Chan, on eTopus's work with Hirose
02 / The long buildFrom one chair to three regions
Chan began building eTopus in 2012. An industry biography written seven years later described the company growing from a one-person operation to an agile team of 30, with a footprint in the United States, Hong Kong and Taiwan. For a semiconductor startup, that first phrase is almost comic: one person facing tape-outs, customers, capital, recruiting and the exquisite indifference of Maxwell's equations.
Yet Chan did not arrive alone in the broader sense. He had already worked through two fabless semiconductor startups and watched both become acquisition stories. He had learned storage and communications, engineering management and the commercial realities around technical work. At eTopus he built with co-founders and specialists including Peter Kou and Danfeng Xu, then extended the network through customers and partners.
The partnerships reveal the strategy. Hirose and Amphenol brought physical interconnect systems to public 112Gbps demonstrations. QuickLogic joined eTopus on a flexible eFPGA chiplet template. In 2024, YorChip selected eTopus technology for a low-latency 200G Ethernet chiplet aimed at edge AI and other timing-sensitive workloads. These are not lifestyle collaborations. Each asks different pieces of the hardware stack to survive one another.
By 2025, eTopus was taking its high-speed interface work to TSMC symposiums in Santa Clara and Taiwan. The company had also moved through a Series B financing chapter and continued to describe its IP as deployed with large networking, storage and 5G customers. Private semiconductor companies rarely offer outsiders a tidy scoreboard. Demonstrations, process nodes, standards support and partner announcements become the visible milestones. Chan's public trail is made of exactly those markers.
Chan's description of leadership is similarly connective. He has said his team is his central source of inspiration, especially engineers working through difficult problems and the sparks produced by cross-functional collaboration. A leader's task, in his account, is not control. It is to create the conditions in which other people can excel.
“No breakthrough happens in isolation.”Harry Chan, on why he revisits The Innovators
03 / An operating systemCuriosity with calluses
The book Chan returns to is Walter Isaacson's The Innovators. He values its account of computing as a convergence of people and disciplines rather than a procession of solitary geniuses. His advice to young engineers follows that reading: keep learning, update your skills, find mentors, build a professional network and remain open to constructive feedback. If you start a company, solve a real problem and make the value clear.
It is sober advice for an industry that can make ambition sound like a substitute for evidence. Semiconductor development rewards conviction, but only after checking it against instruments, standards, customers and manufacturing. Chan's version of persistence has feedback built into it. Fail fast. Learn faster. Continue, but do not continue unchanged.
Stay curious, refresh your skills, seek mentors and invite useful criticism.
Start with a real problem, state the value plainly, and treat resilience as a learning loop.
Then there is running. Chan says it has become part of daily life, although experience has not made the first step automatic. Summer heat and winter cold both offer persuasive arguments for remaining indoors. The contest is at the door. His solution is to begin before readiness arrives and let repetition become an edge.
This is motivational language, but in Chan's story it comes with enough scraped knees to avoid becoming a poster. A semiconductor company can spend years before its work appears inside a customer's chip. A design must pass from model to layout, fabrication to test, demonstration to adoption. Progress sometimes looks like a cleaner eye diagram. Sometimes it looks like the same engineer returning to the same problem with one better question.
04 / The very wide lensNanometers by day, light-years by night
Chan's other hobby stretches the scale in the opposite direction. He is an amateur astronomer. Through a telescope, the light entering his eye may have been traveling for centuries. He finds the experience humbling, but also professionally familiar. Technology asks for microscopic attention to detail and the ability to keep the whole system in view. Astronomy gives him a phrase for the combination: balancing the infinite with the infinitesimal.
It is a useful picture of founder attention. Zoom in far enough and every flaw becomes the universe. Zoom out too far and every plan becomes vague. The job is to change lenses without losing the object. A single noisy link matters; so does the larger shift toward disaggregated computing. One customer test matters; so does an ecosystem of interoperable chiplets.
That larger shift gives Chan's long specialization a new urgency. AI systems consume data movement on an immense scale. Chiplets multiply the number of boundaries inside a package. Faster standards narrow the tolerance for error. The humble interconnect becomes strategic precisely because nobody bought the system merely to admire it. They bought the system to make its parts work together.
Chan has spent his career on that verb: connect. It describes SerDes links, the distributed team he built, the partnerships eTopus uses to test silicon against actual hardware, and the collaborative history he finds in his favorite book. Even his two hobbies fit. Running measures the stubborn distance from here to there. Astronomy begins when a signal finally arrives.
The obvious future is larger models, denser packages and more specialized chips. The less obvious future is the set of tiny roads required to keep them from becoming expensive islands. Harry Chan has been paving those roads for years. He knows that speed is never just a number. It is what survives the journey.