The architectural fact first: capacity and bandwidth pull against each other in memory design, and the practical answer is to use more than one kind of memory at once. US10573368B2, granted February 2020 to Apple (CPC G11C 11/4023), patents exactly that — a system pairing a high-density, low-bandwidth memory with a low-density, high-bandwidth memory.
Gloss the tradeoff once. A wide, fast memory is expensive and hard to make large; a large, cheap memory is slower. Tiering them — keep the hot, bandwidth-hungry data in the fast tier and the bulk in the dense tier — is how systems get both. The same logic now drives AI accelerators that pair scarce HBM with larger pools of standard DRAM.
“In an embodiment, a memory system may include at least two types of DRAM, which differ in at least one characteristic. For example, one DRAM type may be a high density DRAM, while another DRAM type may have lower density but may also have lower latency and higher bandwidth than the first DRAM type.”— U.S. Patent No. 10,573,368 source
Why a moat read cares: Apple is not a merchant memory vendor, but it designs the systems that consume memory, and patents like this protect the integration choices that make its silicon efficient. The competitive value is not in inventing tiering — it is in owning a specific, defensible implementation inside a high-volume product.
The period context sharpens it. In early 2020, memory tiering was an established but pre-AI concern. Reading this grant now shows that the conceptual groundwork for mixing memory classes by bandwidth was already patented years before the AI memory crunch made the tradeoff front-page.
The caveat stands: this is a system-design patent and a defensive asset, not evidence of a shipped configuration or a competitive benchmark. It tells you where Apple staked an integration claim, not how its memory subsystem compares in the field.
For the investor, the durable point is that the memory question was never just "how much." It was always "how much, how fast, at what cost" — and the companies patenting the tiering answer in 2020 were building moat in the layer that now constrains every AI system.