Olefin-Linked Ionic Covalent Organic Frameworks with Carbenium Backbones for High-Performance Neuromorphic Computing

Olefin-linked covalent organic frameworks (ccCOFs) have shown great advantages in the community since discovery. However, the synthesis of ccCOFs highly relies on unsaturated heteroatom-involved segments (uSHS) to activate C─H bond to produce carbanion and to further condense with aldehydes. Such electron-withdrawing uSHS have long hindered the wide applications of ccCOFs in optoelectronic devices due to the strong electron-withdrawing effect. In this work, we report a carbenium-involved synthesis of ccCOFs without uSHS' participating. As-synthesized ionic ccCOFs show good crystallinity and ultra-narrow bandgap down to 1.03 eV without introducing complicated donor-acceptor building blocks. Such ccCOFs can also be directly synthesized as uniform films on different substrates, for example, glass, silicon wafer, Au, and so on. Most impressively, the device Au/2DPPPV-102/Al exhibits 32 distinct conductance states under low-voltage sweeps and each conductance state demonstrated excellent non-volatility and could be maintained stably for over 5000 s, which are typical features of memristor. A convolutional neural network was implemented using a 32-conductance-state memristor array and applied to image recognition, yielding recognition accuracies exceeding 90% for five randomly selected images.