As transistor scaling under traditional Moore’s Law decelerates, the semiconductor industry’s primary innovation engine has shifted toward 3D IC architectures, heterogeneous integration, and hybrid bonding. Stacking dies in Z-space enables extreme interconnect densities for High Bandwidth Memory (HBM) and advanced logic, but it introduces a critical manufacturing bottleneck: detecting nanometer-scale surface variations before bonding occurs. Measuring parameters like copper pad recess across a wafer requires ground-truth 3D precision, a domain long dominated by Atomic Force Microscopy (AFM). Traditional AFMs have been confined to the lab, rather than inline, because they are so slow. A few companies are producing fab AFMs, and speeds have increased over the years.
“AFMs have historically been very slow,” explained David Morris, CEO of Canadian startup ICSPI, in a recent interview with TechSoda. “It’s an amazing technique, but if it’s too slow to provide actionable data, then it doesn’t end up being that useful.”
The Pain Point of Classic Metrology
At the heart of this challenge lies a classic metrology dilemma that has constrained chipmakers for decades. “Historically, you always have this trade-off between throughput and resolution,” Morris pointed out. “If you want to go to finer resolution to see smaller things, you have to slow down.” While optical inspection tools offer high speed, they hit a fundamental physics barrier known as the diffraction limit when trying to resolve features below one micron. Single-probe AFMs deliver sub-nanometer precision, but they are strictly physics-limited by bandwidth, making full-wafer scanning agonizingly slow.
To understand the core technology, Morris offers a clear analogy: an AFM works much like a microscopic record player. Instead of relying on optical lenses, a physical stylus scans across a sample surface to feel atomic forces, capturing exact 3D topographic data down to the sub-nanometer scale. In a conventional lab setup, managing these three-axis movements requires massive equipment. “If you go into a lab and look at a traditional AFM, it’s going to be the size of a refrigerator,” Morris noted. “Most of what is happening in our system is happening on a one-square-millimeter chip—the size of a pinhead.”
Spin-off from University of Waterloo
ICSPI was founded as a spin-out from the University of Waterloo by Dr. Neil Sarkar and Prof. Raafat Mansour to eliminate that massive footprint. By integrating the entire three-axis scanner and force sensors onto a single CMOS-MEMS chip, the company removed the bulky lasers and external piezo stages that make legacy tools fragile and slow.
The real breakthrough for high-volume semiconductor manufacturing, however, lies in massive parallelization. Because a single AFM probe is strictly physics-limited by bandwidth, ICSPI operates arrays of micro-AFMs in parallel to break through the speed barrier.
In public pathfinding collaborations with IMEC presented at SPIE, ICSPI demonstrated scaling from 4 to 10, and recently 20 parallel scan heads operating simultaneously, with an architecture capable of scaling to hundreds or thousands of probes across a full wafer. Operating in parallel allows the platform to deliver atomic-scale 3D Z-axis measurements at - eventually - optical throughput speeds.
Why ICSPI Matters for Advanced Packaging
This capability hits at the exact financial pain point of advanced packaging: catastrophic yield loss during bonding. In hybrid bonding, dielectric surfaces and copper pads must align with nanometer-level spec tolerances. “You’re at the very end of your processing where you have known good dies,” Morris emphasized. “And then you go to bond them together, and your bonding fails, and you lose both of those chips at their most valuable state. Those are the types of costly problems that we’re trying to solve.”
This high-value proposition is bringing major Asian chipmakers to the table. Morris visited Taiwan in late July to meet prospective partners and build relationships across the region’s advanced packaging ecosystem. He highlighted opportunities on two fronts: exploring supply chain partnerships as the company moves toward volume CMOS-MEMS production, and collaborations with Taiwanese equipment makers.
To anchor its presence in the ecosystem, ICSPI is exhibiting for the first time at SEMICON Taiwan, sending a technical delegation to engage local chipmakers, OSATs, and Taiwanese equipment partners.
Having built a mature core technology with several hundred single-chip AFM units already deployed worldwide for R&D, ICSPI has operated leanly on non-dilutive funding from DARPA, NGEN, and Canada’s FABrIC initiative.