The booming artificial intelligence has led to an urgent demand for high-efficient information processing systems. Inspired by the interconnected synapses in human brain, we constructed a novel low-dimensional flexible hybrid photoelectric-modulated artificial heterosynapse with sub-femtojoule energy consumption (0.58 fJ/spike in long-term potentiation (LTP) and 0.86 fJ/spike in long-term depression (LTD)) and ultrafast response (50 ns), which is 10⁵ folds faster than human brain (10 ms). The device synergistically utilizes the remarkable photoelectric properties of a 2D MoSSe channel and a 0D BPQD trap layer to achieve a better computing architecture. The artificial synapse successfully emulates neuromorphic functions under both electric and light stimuli. More importantly, the multi-terminal heterosynaptic plasticity can be modulated effectively by three factors in a cumulative/subtractive way, resembling biological synapses affected by an external neuromodulator, enabling higher order LTP correlations (percentage of increase is ~203% compared with electric modulation) and multiple memory states. The performance of the device was unaffected by substrate bending, indicating the robust stability and high flexibility under mechanical strains. Moreover, the Pavlov’s dog classical conditioning experiments were performed to realize associative learning with the synaptic device. These results highlight a new approach for constructing highly efficient wearable neuromorphic computing systems based on mixed low-dimensional structures.

蓬勃发展的人工智能导致了对高效信息处理系统的迫切需求。受人脑中相互连接的突触的启发，我们构建了一种新型的低维柔性混合光电调制人工异源突触，具有亚飞焦耳的能量消耗（长期增强（LTP）为0.58 fJ / spike，长期增强为0.86 fJ / spike）。项抑制（LTD）和超快响应（50 ns），即10 ⁵折叠速度比人脑快（10毫秒）。该设备协同利用2D MoSSe通道和0D BPQD陷阱层的卓越光电性能来实现更好的计算架构。人工突触成功地模拟了电刺激和光刺激下的神经形态功能。更重要的是，可以通过三个因素以累加/减法方式有效地调节多末端异突触可塑性，类似于受外部神经调节剂影响的生物突触，实现更高阶的LTP相关性（与电调节相比，增加百分比为〜203％）。和多种记忆状态。器件的性能不受衬底弯曲的影响，表明在机械应变下的鲁棒稳定性和高柔韧性。此外，进行了巴甫洛夫犬的经典条件实验，以实现与突触装置的联想学习。这些结果凸显了一种基于混合低维结构构建高效可穿戴神经形态计算系统的新方法。