Fig. 2.

Reconfigurable Gaussian synapse. a Schematic of a reconfigurable Gaussian synapse involving dual-gated n-type MoS₂ FET and p-type BP FET. The top-gate stack was fabricated using hydrogen silsesquioxane (HSQ) as the top-gate dielectric and nickel/gold (Ni/Au) as the top-gate electrode. b Back-gated transfer characteristics of the MoS₂ FET at $V_{\mathrm{D}}$ = 1 V, for different top-gate voltages ($V_{\mathrm{N}}$). c Back-gate threshold voltage, $V_{\mathrm{TN}}$ of MoS₂ FET as function of $V_{\mathrm{N}}$ was extracted using the constant current method. Inset shows the band diagram elucidating how $V_{\mathrm{N}}$ controls $V_{\mathrm{TN}}$ by electrostatically adjusting the height of the thermal energy barrier for electron injection into the MoS₂ channel. d Back-gated transfer characteristics of the BP FET at $V_{\mathrm{D}}$ = 1 V, for different top-gate voltages ($V_{\mathrm{P}}$). e Back-gate threshold voltage, $V_{\mathrm{TP}}$ of BP FET as function of $V_{\mathrm{P}}$. Inset shows the band diagram elucidating how $V_{\mathrm{P}}$ controls $V_{\mathrm{TP}}$ by electrostatically adjusting the height of the thermal energy barrier for hole injection into the BP channel. As expected the slope (${\alpha }_{\mathrm{N}}=1.91$) of $V_{\mathrm{TN}}$ versus $V_{\mathrm{N}}$ and the slope (${\alpha }_{\mathrm{P}}=2$) of $V_{\mathrm{TP}}$ versus $V_{\mathrm{P}}$ are found to be similar and equal to the ratio of top-gate capacitance ($C_{\mathrm{TG}}$) to the back-gate capacitance ($C_{\mathrm{BG}}$), which in our case was ≈1.94. f Transfer characteristics of the Gaussian synapse for different values of $V_{\mathrm{N}}=V_{\mathrm{P}}$. This configuration allows us to shift the mean (${\mu }_{\mathrm{V}}$) of the Gaussian synapse without changing the amplitude (A) and the standard deviation (${\sigma }_{\mathrm{V}}$). g ${\mu }_{\mathrm{V}}$ as a function of $V_{\mathrm{N}}=V_{\mathrm{P}}$. h Transfer characteristics of the Gaussian synapse for different values of $V_{\mathrm{N}}=-V_{\mathrm{P}}$. This configuration allows us to configure ${\sigma }_{\mathrm{V}}$ while keeping ${\mu }_{\mathrm{V}}$ constant. i ${\sigma }_{\mathrm{V}}$ as a function of $V_{\mathrm{N}}=-V_{\mathrm{P}}$. However, this configuration also results in an increase in the amplitude (A) of the Gaussian synapse as ${\sigma }_{\mathrm{V}}$ increases. This increase can be adjusted by changing the drain voltage ($V_{\mathrm{D}}$) since A is linearly proportional to $V_{\mathrm{D}}$. Nevertheless, by controlling $V_{\mathrm{N}},V_{\mathrm{P}}$ and V_D, it is possible to adjust the mean, standard deviation and amplitude of the Gaussian synapse