Researchers at University of Edinburgh, led by Xiangyu Ren, have developed a new compilation scheme to enhance the reliability of photonic quantum computing. The advancement directly addresses a critical limitation within measurement-based quantum computation (MBQC), where probabilistic fusion operations are inherently vulnerable to both fusion failure and fusion erasure errors. Photon loss, a pervasive issue leading to fusion errors, can be substantially mitigated through optimised compilation strategies and the innovative use of quantum memory. Tree-encoded fusion surpasses photon loss limitations in silicon spin qubit systems A two-qubit fusion fidelity of 99.72% has been achieved, significantly exceeding a key threshold that previously constrained the scalability of photonic quantum computation. This level of accuracy, coupled with a silicon-based spin qubit quantum memory, facilitates the generation of more complex graph states than was previously attainable. Prior to this development, photon loss during fusion operations severely restricted the reliability and complexity of quantum programs. The newly developed compilation scheme, incorporating ‘tree-encoded fusion’, actively mitigates fusion erasure errors, which arise from lost photons, a problem that has been largely unaddressed by existing compilers such as OneAdapt. The significance of achieving such high fidelity lies in its potential to unlock more intricate quantum algorithms and larger-scale quantum processors. Six quantum algorithm benchmarks revealed an exponential performance improvement when compared to the OneAdapt compiler, a leading tool for all-photonic architectures. This substantial improvement highlights the efficacy of the tree-encoded fusion approach in optimising quantum program execution. Using a silicon-based spin qubit quantum memory, validation on real photonic quantum computing hardware confirmed the feasibility of this approach, successfully generating caterpillar states for efficient entanglement distribution. Caterpillar states are particularly useful as they represent a foundational resource for various quantum algorithms. This work builds upon initial successes by detailing the underlying mechanism for improved performance, moving beyond demonstrating improvement