The moat fact first: the chiplet era created a new contested layer — how dies and memory connect at high bandwidth across packages — and startups are racing to own IP in it. US12058874B1, granted August 2024 to Eliyan, patents a universal network-attached memory architecture (CPC H10B 80/00).
Gloss it once. As systems move from single monolithic chips to many chiplets, the connections between them — die-to-die and to pools of memory — become a defining bottleneck. A "network-attached memory" architecture treats memory as a resource reachable over a high-bandwidth interconnect fabric, rather than hard-wired to one processor. The IP is in how that fabric is structured.
Why a moat read cares: the big systems vendors need interconnect IP, and they can either build it or license it. A focused startup holding foundational architecture patents in this layer has a licensing-and-acquisition thesis: own the standard-adjacent IP that everyone building chiplet systems eventually needs.
The period framing matters. By 2024, chiplet interconnect (including the UCIe standardization effort) was a hot frontier, and capital was flowing to interconnect specialists. A startup grant at that moment reflects the land-grab for IP in the die-to-die and memory-fabric layer.
The caveat we attach: this is an architecture patent and a defensive/licensing asset. It evidences Eliyan's IP position; it does not establish adoption, licensing revenue, or that any large vendor depends on it.
For the period investor, the durable point is that interconnect became a venture-scale opportunity. A 2024 network-attached-memory grant from a startup is a marker of where small players are betting they can own a layer the giants need.