Transistor-based circuits with parallel computing architectures and distributed memories, such as graphics processing units (GPUs) from Nvidia, [9] tensor processing units (TPUs) from Google, [3,10] field-programmable gate arrays (FPGAs) from Intel, [11] and the TrueNorth neuromorphic circuit from IBM [12] have been developed to improve their energy efficiencies ( Figure 1a) to the range of 10 10 − 10 11 FLOPS W −1 (floating point operations per second per watt) by increasing parallelism and reducing global data transmission. However, their energy efficiencies are fundamentally limited by the energy consumptions on memory (≈10 −15 J bit −1 ) and signal transitions (≈10 −11 J bit −1 ) in digital computing circuits. [5,6] When transistors approach the limitations of their minimal sizes near the end of Moore's law, the energy efficiencies of transistorbased computing circuits are asymptotically saturated [4][5][6]13,14] (Figure 1a). Meanwhile, the information industry generates "big data" with exponentially increasing volumes, and leads to exponentially increasing power requirements for computations. [4,6,14] This trajectory is unsustainable as it would exceed the entire global power production in one or two decades [15] (Figure 1a). It is imperative to develop a new platform to facilitate inference and learning from "big data" in emerging intelligent systems with significantly higher energy efficiency than that of the transistor-based Turing computing platform.The human brain performs inference and learning from "big data" with an estimated speed (≈10 16 FLOPS) [16] comparable to the speed (≈10 17 FLOPS) of the fastest supercomputer, Summit, [17] but consumes much less power (≈20 W) than the supercomputer (≈10 7 W), and is much more energy-efficient (≈10 15 FLOPS W −1 ) than the supercomputer (≈10 10 FLOPS W −1 , Figure 1a). By contrast, the human brain concurrently performs spatiotemporal inference and learning in analog parallel mode [16,18,19] (Figure 1c) via a network of neurons connected by ≈10 14 synapses (Figure 1d). For inference, a wave of voltage pulses, V t m ( ) i Adv.

Unlike artificial intelligent systems based on computers which have to be programmed for specific tasks, the human brain “self‐programs” in real time to create new tactics and adapt to arbitrary environments. Computers embedded in artificial intelligent systems execute arbitrary signal‐processing algorithms to outperform humans at specific tasks, but without the real‐time self‐programming functionality, they are preprogrammed by humans, fail in unpredictable environments beyond their preprogrammed domains, and lack general intelligence in arbitrary environments. Herein, a synaptic resistor circuit that self‐programs in arbitrary and unpredictable environments in real time is demonstrated. By integrating the synaptic signal processing, memory, and correlative learning functions in each synaptic resistor, the synaptic resistor circuit processes signals and self‐programs the circuit concurrently in real time with an energy efficiency about six orders higher than those of computers. In comparison with humans and a preprogrammed computer, the self‐programming synaptic resistor circuit dynamically modifies its algorithm to control a morphing wing in an unpredictable aerodynamic environment to improve its performance function with superior self‐programming speeds and accuracy. The synaptic resistor circuits potentially circumvent the fundamental limitations of computers, leading to a new intelligent platform with real‐time self‐programming functionality for artificial general intelligence.

Unlike artificial intelligent systems based on computers, which must be programmed for specific tasks, the human brain can learn in real-time to create new tactics and adapt to complex, unpredictable environments. Computers embedded in artificial intelligent systems can execute arbitrary inference algorithms capable of outperforming humans at specific tasks. However, without real-time self-programming functionality, they must be preprogrammed by humans and will likely to fail in unpredictable environments beyond their preprogrammed domains. In this work, a Si-based synaptic resistor (synstor) was developed by integrating Al2Ox/TaOy materials to emulate biological synapses. The synstors were characterized, and their operation mechanism based on the charge stored in the oxygen vacancies in the Al2Ox material was simulated and analyzed, to understand the inference, learning, and memory functions of the synstors. A self-programming neuromorphic integrated circuit (SNIC) based on synstors was fabricated to execute inference and learning algorithms concurrently in real-time with an energy efficiency more than six-orders of magnitudes higher than those of standard digital computers. The SNIC dynamically modified its algorithm in a real-time learning process to control a morphing wing, thus successfully improving its lift-to-drag force ratio and recovering the wing from stall in complex aerodynamic environments. The synaptic resistor circuits can potentially circumvent the fundamental limitations of computers, thus providing a platform analogous to neurobiological network with real-time self-programming functionality for artificial intelligent systems.