Computational dissipation may arise from both information erasure and finite-time operation, although both contributions can in principle be avoided in the absence of space and time constraints. Logically reversible computation can replace erasure with ancilla cleanup, and slow adiabatic protocols can eliminate any excess finite-time dissipation. Under limited memory and in finite time, however, choosing reversible cleanup over erasure requires additional dissipative operations, creating a semi-reversible energy--memory--time tradeoff. We formulate this tradeoff at a fixed operating point of an externally clocked thermally coupled computer, where reversible checkpoint moves and local irreversible erasures have specified effective prices $ε_R$ and $ε_I$. Their ratio $λ=ε_I/ε_R$ selects an energy-optimal semi-reversible schedule under a checkpoint-memory bound by trading reversible moves against erasures. For a sequential computation whose dependency graph is a line, we derive an exact separator recurrence and prove that an asymptotically optimal schedule repeats a strictly reversible block followed by one local erasure. Coupling this scheduling theorem to a stochastic-thermodynamic constitutive model shows how temperature, primitive duration, memory, and device-dependent operation costs determine whether the optimal schedule uses reversible cleanup, elective erasures, or erasure at every disposal.

翻译：暂无翻译