Neuromorphic computing, which mimics biological neural networks, can overcome the high-power and large-throughput problems of current von Neumann computing. Two-terminal memristors are regarded as promising candidates for artificial synapses, which are the fundamental functional units of neuromorphic computing systems. All-inorganic CsPbI 3 perovskite-based memristors are feasible to use in resistive switching memory and artificial synapses due to their fast ion migration. However, the ideal perovskite phase α-CsPbI 3 is structurally unstable at ambient temperature and rapidly degrades to a non-perovskite δ-CsPbI 3 phase. Here, dual-phase (Cs 3 Bi 2 I 9 ) 0.4 −(CsPbI 3 ) 0.6 is successfully fabricated to achieve improved air stability and surface morphology compared to each single phase. Notably, the Ag/polymethylmethacrylate/(Cs 3 Bi 2 I 9 ) 0.4 −(CsPbI 3 ) 0.6 /Pt device exhibits non-volatile memory functions with an endurance of ≈10 3 cycles and retention of ≈10 4 s with low operation voltages. Moreover, the device successfully emulates synaptic behavior such as long-term potentiation/depression and spike timing/widthdependent plasticity. This study will contribute to improving the structural and mechanical stability of all-inorganic halide perovskites (IHPs) via the formation of dual phase. In addition, it proves the great potential of IHPs for use in low-power non-volatile memory devices and electronic synapses.The ORCID identification number(s) for the author(s) of this article can be found under https://doi.org/10.1002/adfm.201906686.femto-Joule scale. [1,2] Neuromorphic computing systems that emulate the human brain are considered promising candidates for overcoming the energy and throughput limitations of conventional von Neumann computing systems. Inspired by the human brain, neuromorphic computing systems are composed of electronic neurons and synapses, and achieve a high degree of parallelism with the physically united memory and information processing unit. Artificial synapses are fundamental functional units of neuromorphic architectures, where electrical stimuli from a pre-neuron are transmitted to a post-neuron, thereby generating updated synaptic weight (i.e., causing a conductance change in an electronic device). Notably, resistive switching (RS) memory has been actively investigated for synaptic devices due to its scalability, simple structure, fast operation speed, and low-energy consumption, which are the most important requirements for neuromorphic computing. [3,4] In the last few years, halide perovskites (HPs) with the chemical formula ABX 3 [where A is an organic or inorganic (Cs or Rb) cation, B is a divalent metal cation (Pb or Sn), and X is a halide anion (I, Cl, or Br)] have been widely investigated for RS memory and artificial synapses because of their tunable bandgaps and fast ion migration. [5][6][7] To date, organolead HPs (OHPs, A = methylammonium (MA), B = Pb, and X = I, Cl, or Br) have been successfully utilized as the active layer of memristor-based artificial synapses, emulat...