Multi-input　synaptic　devices　that　can　imitate　multi-synaptic　connection　and　integration　in　the　human　brain　are　crucial　for　the　construction　of　ideal　brain-like　computing　systems　with　parallelism,　low　power　consumption,　and　robustness.　However,　the　current　multi-input　synaptic　devices　are　all　based　on　electrical　operation,　bringing　undesirable　signal　redundancy,　huge　device　scale,　and　large　energy　consumption.　Here,　for　the　first　time,　a　multi-input　synaptic　device　utilizing　light　as　an　input　signal　was　realized　to　overcome　the　drawbacks　induced　by　electrical　operation.　The　essential　synaptic　functions　including　synaptic　short-term　plasticity　(STP),　long-term　plasticity　(LTP),　and　paired-pulse　facilitation　(PPF)　were　demonstrated　in　an　energy-efficient　way　(picojoule　magnitude)　and　synaptic　superlinear/sublinear　integrations　were　successfully　imitated　in　our　device.　Moreover,　the　＂AND＂/＂OR＂　light　logic　functions　and　the　light-stimulated　Pavlov　classical　conditioning　experiment　with　dogs　were　realized　based　on　the　synaptic　integration　behaviors.　Interestingly,　owing　to　the　persistent　photoconductivity　(PPC)　effect　of　the　semiconductor　layer　and　synaptic　behaviors　induced　by　perovskite　quantum　dot　(QD)　absorption,　the　device　could　implement　detection,　acquisition,　analysis,　and　storage　of　light　signals,　enabling　its　potential　application　in　fog　computing.　This　powerful　device　offers　the　potential　for　building　an　artificial　intelligence　neural　network　with　miniaturization,　low　energy　consumption,　and　superior　connectivity　between　discrete　computing　modules.　This　journal　is　©　The　Royal　Society　of　Chemistry.