In this study, an artificial synapse semiconductor device of neuromorphic engineering was manufactured using flash architecture by simulating the signal processing method of the human brain, that is, the behavior of neurons and
synapses, and this devi...
In this study, an artificial synapse semiconductor device of neuromorphic engineering was manufactured using flash architecture by simulating the signal processing method of the human brain, that is, the behavior of neurons and
synapses, and this device was measured electrically. We employed a gate stack structure of Pt/Cr/HfO2/Pt/Cr/HfO2/SiO2/Si to represent the signal processing of synapses and neurons. The floating gate (FG) Pt was fabricated in the form of a mesh island to reduce the leakage current of the device and to take advantage of the device yield stability. The thickness of Pt and Cr deposited on the thin film on top was measured to enable electron tunneling to and from the Pt, surrounded by metal and HfO2 insulating layers, altering the threshold voltage (VT) of the device and setting the synaptic weight in an analog form by storing or erasing electrons from Si. Voltage was applied to the Control Gate of the device for program and erase, representing the weakening and strengthening of synaptic connection strength. To prevent insulation breakdown and damage to the device during operation, an electrically measured Current-Voltage across the Tunnel Oxide HfO2/SiO2 confirmed that ±7 V or lower voltage allowed electron tunneling without insulation breakdown. The thickness of the Control Oxide was adjusted to control the memory window width, allowing for a wider range of synaptic weight division. The artificial synaptic device's synaptic connection strength was expressed by iteratively programming and erasing, and the endurance of the device was confirmed as part of its reliability. To determine the appropriate voltage applied to the Control Gate, considering the voltage range that may cause damage or leakage current during device operation, the synaptic weight and device endurance were taken into account. Additionally, the device's retention was verified, ensuring its suitability for charge storage, and passivation experiments were conducted to enhance the device's retention. Finally, by increasing the density of FG, we manufactured a device with extended
capacitance, memory window, and endurance. It is anticipated that optimizing the insulation layer and metal thickness will further enhance the performance, making it a potential candidate for a highly improved single artificial synaptic device.
Keywords: Neuromorphic, threshold voltage, High-k, Charge storage