IHW Patent Portfolio Analysis

The money arrived five years after the filings

A $37.1M Series A from Korea Development Bank, Mirae Asset Venture Investment and IBK Capital. Korean policy capital, Korean institutional capital, Korean bank capital. The kind of syndicate that shows up when a semiconductor thesis has already been checked by people who understand fabs. IHW was founded in 2021. The company that raised this round is four years old.

The patent record is older than the company. The first two families were filed in 2020, before incorporation. That gap is the most interesting number in the brief, because it means the technical position predates the corporate one. Somebody had made the decision about what to claim before there was an entity to hold it, and before there was a round to fund building it out.

The patent record is older than the company.

Four ticks, two of them solid

Four families since 2020. Two granted and enforceable today. Two published and pending, with the first of those close enough to grant that it should be read as a near-term asset rather than a hope. Filings sit in CN, EP, KR and US, which is the full set of jurisdictions that matter for a memory-adjacent chip business and an expensive choice for a company this size.

The velocity is lumpy: two in 2020, one in 2022, one in 2025. That is not a machine producing filings on a schedule. It is a small group filing when something specific got solved. The titles say what: an analog computing unit for representing negative weights, an analog-digital hybrid computing method and neuromorphic system, and a method of initializing a memory cell containing chalcogen compound.

Jun-Sung Kim is on three of four

Four names across the whole portfolio. Jun-Sung Kim appears on three families, Sang-Hoon Yoon on two, Hwan Jun Zang on one. This is not a research organisation with a patent committee. It is a handful of people, and one of them is on nearly everything that has been filed since 2020, including the 2020 work that predates the company.

Concentration like this cuts two ways and it is worth being honest about both. A portfolio where one inventor carries three of four families has a coherence that larger ones rarely achieve: the claims are likely to fit together, because the same person wrote the problem statement each time. It also means the institutional memory of why each claim is drafted the way it is lives in a very small number of heads.

Samsung has filed here two thousand times

The claims cluster in G11C, memory, and G06N, AI and machine learning. Over the last five years the same CPC groups drew 2,048 families from Samsung, 1,142 from Micron, 961 from Huawei, 870 from IBM, 715 from Google and 637 from Bosch Robert. Against that, four. On a drawn-to-scale diagram, IHW is a dot next to a set of planets.

Count is the wrong measure here and the crowd proves it. In a space this dense, nobody wins on volume, because the volume is already taken. What a four-family portfolio can do is claim something narrow that the large filers did not think to claim, and hold it. The relevant question is not how many patents IHW has. It is how specific the granted claims are.

In a space this dense, nobody wins on volume, because the volume is already taken.

Memory manufacturers are reading the file

Five organisations cite this portfolio. Intelligent Hw twice, Yangtze Memory twice, Seoul Nat Univ R&Db Foundation once, Korea Inst Sci & once, SK Hynix once. Look at what those names have in common. Yangtze Memory and SK Hynix build memory at scale. The two research institutions are Korean, and sit exactly where this kind of device physics gets worked out before it reaches a production line.

Citations from a memory manufacturer to a four-family AI chip portfolio tell you where the claims actually landed. Not in the model layer. In the cell. The chalcogen compound initialization filing and the analog unit representing negative weights are both storage-device claims dressed as compute, and the people citing them are the people who make storage devices.

What the next filing has to say

IHW describes itself as a builder of energy-efficient AI chips that run inference on devices. The portfolio describes something narrower and more defensible: a way of doing arithmetic inside memory cells, claimed in four jurisdictions, by a group small enough to fit around one table. The single open role is an SoC/IP Design Engineer, which is a building hire, not a research one.

The 2025 filing matters more than the three before it, because it is the first one made with $37.1M behind it. Whether it continues the memory-cell thread or opens a second one will say more about the company's next five years than the round did. A same-week neighbour in G06N3 was Etched, which tells you how contested the neighbourhood has become.