Polypills are single oral dosage forms that combine multiple active pharmaceutical ingredients and excipients, enabling fixed-dose combination therapies and tunable multi-phase release. Recent advances in additive manufacturing make it possible to fabricate heterogeneous, multi-material tablets with tailored internal architectures; however, current formulations are often tuned via ad hoc parameter sweeps, limiting systematic exploration of the coupled design space of shape, composition, and dissolution dynamics. We present PILLTOP, a multi-material topology optimization framework that co-designs polypill geometry and excipient distributions to match prescribed drug-release kinetics under physically resolved dissolution-diffusion behavior. The forward model couples a modified Allen-Cahn phase-field equation for interface evolution with Fickian diffusion for solute transport, and the target is posed in terms of matching a prescribed instantaneous mass release-rate profile. To balance manufacturable, connected topologies with high design freedom, we use a hybrid parameterization in which supershapes define the global pill shape while a coordinate-based neural network encodes spatially varying excipient distributions; discrete-material realizability and constituent usage are enforced via grayness suppression and minimum volume-fraction constraints. The framework is implemented end-to-end in JAX, leveraging automatic differentiation to compute sensitivities through the transient, nonlinear solver for gradient-based optimization. We demonstrate that single-material geometric optimization is insufficient for complex non-monotonic targets, whereas multi-excipient co-optimization accurately tracks both monotonic and non-monotonic release-rate profiles. Finally, we illustrate degradation-aware design inspired by pharmaceutical aging data by optimizing excipient allocations across feedstock grades under minimum-usage constraints, enabling close matching of target kinetics despite degraded dissolution performance.