This paper examines whether the elimination of volatility hierarchies can be sustained across performance and cost regimes that have historically justified hierarchical memory systems. Using a unified semiconductor memory architecture as a fixed reference, we analyze read latency, write behavior, retention stability, access uniformity, and cost scaling relative to fully realized DRAM–NAND systems. Rather than relying on device-specific benchmarks, the analysis identifies the dominant structural constraints that bound each performance domain. We show that fast, non-destructive reads, localized and deterministic writes, long-term retention without refresh, and uniform access semantics can coexist without performance degradation. Cost per bit is shown to follow a favorable scaling trajectory driven by density and integration rather than system-level overhead. These results remove the remaining technical and economic rationale for maintaining hierarchical memory designs.
Anonymous (Tue,) studied this question.