No products in the cart.
Chen, Q., Zhang, Y., Liu, S., Han, T., Chen, X., Xu, Y., Meng, Z., Zhang, G., Zheng, X., Zhao, J., Cao, G., Liu, G.
Advanced Intelligent Systems
PiFM images of the CsFAMA film, sequentially recorded in the a) pristine state, b–f) after being subjected to voltage stressing with the amplitude of −1 to −0.2 V and i–n) after being subjected to voltage stressing with the amplitude of 0 to 1 V and a ramping step of 0.2 V, respectively. g,h) The operations of perovskite photovoltaic sensory devices under positive and negative voltage stressing conditions. SV and RV represent the stressing and the reading voltages, respectively. The red spheres, yellow spheres, and the green octahedrons inside the perovskite layer represent the cationic FA+/MA+ species, anionic I−/Br− species and anionic PbI2.55Br0.45− of CsFAMA, respectively. The migrated cationic FA+/MA+ species and anionic I‐ species are not drawn to scale for presenting clearance purpose. Conventional artificial visual system relies on the complementary metal‐oxide semiconductor (CMOS)‐based image sensor and field‐programmable gate array (FPGA) to convert the optical signal into voltage spikes. An additional neural network is required to execute convolution for target recognition, which is recently intensively performed with memristor crossbar arrays. Machine vision involving both the self‐adaptive image sensing and in‐sensor computing may enable smart and faithful capturing of the visual information from the environment, as well as offer in situ efficient image‐processing capability by avoiding the massive shuttling of abundant data between the sensor and computing units.[20, 29] This is critically important for user‐end applications wherein the fast responding and decision‐making is vital, e.g., instant detection and dodging of on‐road obstacle during vehicle autopiloting. MAC is the core module of convolution for target recognition in ANNs. Crossbar architecture and parallel operation of the interconnected sensors may allow the natural summation of the device currents following Kirchhoff's law, whereas multiplication can be realized with Ohm's law of _I = G·V_ with _G_ donating the device's conductance. In the present CsFAMA devices, the current–voltage characteristics show strong dependence on optical illumination (Figure 3e). Replotting the device conductance and photocurrent against the irradiation intensity _P_ gives linear relationships of _G = a·P + G_0 and _I = aV·P + I_0_ = R·P + I_0 (or _I _− _I_0_ = aV·P = R·P_) that satisfy the multiplication requirements of neural networks, where _a_ is a prefactor constant, _G_0 and _I_0 are the device conductance and current under dark condition, and _R = aV_ mathematically equals to the photoresponsivity of the system (Figure 3f). It is noteworthy that the photoresponsivity of the CsFAMA devices is a functional of the perovskite's internal ionic state and can be controlled by the biased voltage _V_ nonvolatilely as demonstrated previously, allowing itself as linearly updatable synaptic weights to optoelectronically train the perovskite photovoltaic sensory array for target‐recognition tasks without additional neural network hardwares.