Neuromorphic Generative Design: Brain-Inspired Computing Workflows for Self-Adaptive Building Façades

Introduction
Building envelopes that learn from their environment could transform energy performance, yet conventional rule-based controllers cannot cope with the non-linear, multi-objective nature of façade behaviour. Neuromorphic computing—hardware that reproduces spike-based brain signalling—offers ultra-low-power, real-time inference. This study presents the first closed-loop workflow in which a spiking neural network (SNN) continuously generates and retrains timber-CLT shading morphologies in response to live sensor streams.

Methods
A three-layer SNN comprising 512 leaky-integrate-and-fire neurons was synthesised on a 28 nm field-programmable gate array (FPGA). Photometric, surface-temperature, and occupancy data (1 Hz) from a living-lab office in Lusaka were latency-encoded into spike trains. Network outputs drove a Grasshopper script that reconfigured 240 modular louvred panels; performance feedback—daylight autonomy, glare probability, and predicted mean vote—was returned via EnergyPlus-Radiance co-simulation within a 10 s control horizon. A proportional–integral–derivative (PID) façade served as a benchmark.

Results
After a four-day online-learning phase, the neuromorphic façade reduced annual cooling loads by 35 %, heating by 9 %, and electric-lighting energy by 27 % versus the static baseline, while maintaining 95 % of occupied hours within ASHRAE-55 comfort limits. Decision latency averaged 7 ms and FPGA power draw was only 2.3 W—an 86 % saving over a GPU-based SNN. Actuator travel distance fell by 42 % compared with the PID system, indicating longer mechanical life. A cradle-to-gate life-cycle assessment revealed a 12 % reduction in embodied carbon owing to material right-sizing enabled by the adaptive logic.

Conclusions
Brain-inspired generative design allows building skins to self-optimize luminous, thermal, and energy variables with negligible computational overhead. Coupling neuromorphic hardware with parametric timber components charts a scalable pathway toward self-learning, net-zero-ready façades across diverse climates.