Mechanical metamaterials (MMs) receive widespread attention due to their unprecedented mechanical properties. However, in the next-generation MMs, the cognitive function of information processing can be realized while maintaining superior mechanical properties. Herein, a mechanical metamaterial-based self-powered electronic skin (e-skin) with multimodal fusion perception capability and shape memory reconfigurability is proposed. Benefiting from an MM skeleton and its analytical model, e-skin realizes biomimetic nonlinear mechanical behavior and mechanical reconfigurability to imitate target biotissues. Its integrated perovskite-based elastic sensors enable high-precision collection of physiological movements and auditory, tactile, and precontact distance signals. Further, by imitating the integration and interaction functions in biological multisensory neural networks, the system achieves advanced cognitive functions of acquiring, identifying, and integrating information across modalities. Applications of e-skin are demonstrated in motion monitoring, speech recognition, and somatosensory game operation. These capabilities can be applied to cross-modal perception robot systems based on multisensory neural networks.