Perovskite based solar cells have seen a remarkable increase in power conversion efficiencies (PCEs) since their advent about a decade ago. Both single junction and tandem devices in conjunction with silicon are among the most successful technologies within the research field of photovoltaics. In this work, we are particularly interested in the latter category where PCEs of >30% are expected to be achieved in the near future. Thus, the Shockley-Queisser-limit of a single-junction device can be overcome by using this technology. Combining two semiconductors into a tandem device is usually achieved in one of two ways: either in a monolithic, series-connected design with a recombination contact between the subcells (two-terminal or 2T), or as individually contacted, mechanically stacked subcells (four-terminal or 4T). While 4T tandems have the advantage of electrically independent subcells, they face a potentially more complex fabrication route and module integration necessitating two MPP-trackers. A simpler device design, in turn, makes 2T tandems especially attractive. This advantage is, however, offset by a requirement for current-matched subcells, which cannot be guaranteed at all times outside a laboratory environment and might lower the annual energy yield of this technology. As an alternative to these more common approaches, we investigate three-terminal tandem solar cells (3T) where both subcells share either an electron or a hole contact and each subcell features a separate contact for the other charge carrier species, thereby combining advantages of 2T and 4T tandems. In such 3T devices, different interconnection schemes (depending on the shared contact, bottom cell doping types, and inter-subcell junction) can lead to different electrical behaviours. Here, we focus on a 3T perovskite/silicon heterojunction tandem device with interdigitated back-contacts and a p/n recombination junction between subcells, which leads to a quasi-2T configuration with an additional third terminal. This additional contact can be used to inject or extract surplus current into or from the bottom cell as needed. This leads to a device that is very robust against different spectral conditions and, despite being otherwise quite similar to a 2T device, does not require current matched subcells. This is proven by adjusting the LED intensity of our solar simulator in the wavelength regime of 800–900 nm (and thereby photogeneration in the bottom cell), thereby intentionally mismatching the subcells. We have achieved a combined PCE of 23.3% under AM1.5g conditions. The performance is still limited by reflection losses in the near-IR and a not fully optimised light-incoupling scheme. Future advances in experimentation will overcome these issues and close the gap between performances of record 2T and 3T devices.