CEO Interview with Dr. Albert Liu of Kneron

**Skyler Monroe** (0:10)
Hey, everyone, welcome back to the AI Hardware Show. I'm your host Skyler Monroe, and this is the podcast where we dig into the chips, silicon and infrastructure powering the AI revolution. Big thanks to our sponsors, Limitless AI, helping businesses actually put AI to work in their day-to-day workflows, and a go consulting, that's AGO, your go-to team for silicon development consulting.
Today, we've got a packed episode. We're talking edge AI with a fascinating CEO interview, TSMC transforming rice fields into advanced packaging hubs, South Korea cashing in on the AI chip boom, and SK Hynex making a massive splash on NASDAQ. Let's get into it. All right, kicking things off with a really interesting one, a CEO interview over at SemiWiki, featuring Dr. Albert Liu, the founder and CEO of Kneron. Now, if you haven't heard of Kneron, buckle up because these guys are doing something genuinely compelling in the edge AI space. Dr. Liu is described as a pioneer in NPU architecture that's neural processing unit, and he spent decades working on AI hardware, semiconductor systems and computer vision. So this is not someone who just stumbled into the AI hype cycle. This is a guy who's been in the trenches building this stuff for a long time. So let's set the stage. What is Kneron actually doing? They're a full stack edge AI company. And when I say full stack, I mean, they're not just designing chips, they're building the entire solution from the silicon all the way up to the software and the AI models that run on it. That's a big deal, because a lot of companies in this space either focus on the hardware side or the software side, but doing both well is really hard and it creates a much stickier product. Now the NPU piece is what I really want to unpack here. We talk a lot on this show about GPOs and their dominance in AI training. But when you move to the edge, think cameras, smartphones, IoT devices, industrial sensors. You can't just strap a data center GPU to a security camera. You need something purpose-built for inference, something that's power efficient, small and fast. That's exactly what an NPU is designed to do. Think of it this way. AGPU is like a massive freight train, incredibly powerful, moves enormous amounts of data, but it needs a huge amount of fuel and infrastructure to operate. An NPU is more like a precision delivery drone, smaller, way more efficient for specific tasks, and it can operate independently without being tethered to some massive power grid. For edge applications, that drone wins every single time. What makes Kneron interesting is their focus on what Dr. Liu calls localized AI infrastructure. This is a concept that's gaining a lot of traction, essentially, the idea that not everything needs to go to the cloud for processing. Privacy concerns, latency requirements, and bandwidth costs are all pushing compute closer to where data is actually generated. And Kneron is betting big that the future of AI is distributed, not centralized. The implications here are pretty significant. As AI becomes embedded in more physical devices, your car, your home security system, your factory floor, the demand for efficient edge, AI chips is going to explode. Kneron is positioning itself as a key player in that wave. And with Dr. Liu's background in both the hardware architecture and the systems level, they've got a really strong technical foundation to build on. Definitely one to keep an eye on. Check out the full interview through the link in show notes. It's a great read. Okay, story number two, and this one is fascinating on multiple levels. TSMC, the world's most important chipmaker, is transforming an area of Chia-Yi in Taiwan that was literally rice fields not long ago into what could become Taiwan's next major advanced packaging hub. This comes from the folks over at Semivision, and it's a story about geography, geopolitics and the absolutely relentless demand for AI chips, all rolled into one. So first, let's talk about why packaging matters so much right now. For a long time, semiconductor packaging was seen as kind of the boring cousin of chip fabrication. The sexy work was in the fab shrinking transistors, pushing process nodes forward. But packaging has had a complete renaissance in the AI era. And it's because of one core challenge, getting data between chips fast enough. Here's an analogy. Imagine you've got the world's fastest chef. That's your compute chip. But to cook, they need ingredients delivered from a warehouse. That's your memory. If the delivery road between the warehouse and the kitchen is narrow and slow, it doesn't matter how fast the chef is.
Advanced packaging is essentially about building a superhighway between those chips so data flows as fast as physically possible.