• capacitive transimpedance amplifier
• 28nm CMOS Technology
• neural networks
• deep learning
• analog front-end

The aim of this master's thesis is the analysis and design of an analog front-end for electro-photonic neural networks in Technology TSMC 28 nm with a nominal supply voltage of 0.9 V.
The integrated circuit consists of a capacitive transimpedance amplifier (CTIA), a current-voltage converter able to accumulate the charge obtained by integrating the photocurrent (which is the result of multiplication that took place in photonics) generated by the doubly balanced photodetectors.
The integrating front-end allows the accumulation in the analog domain the results of several operations before sampling, relaxing the ADC bandwidth. Also, an inverting amplifier connected to the output node of the CTIA was used to realize a differential output for a correct conversation with the ADC.
Analysis and design of four different amplifiers, inserted in the CTIA, joined by the same inverter-based architecture is presented. The pre-layout performance in terms of DC gain, bandwidth, GBW and settling time was evaluated, leading to the optimal choice (cascode inverter based amplifier).
Finally, from the transient simulations, the circuit proves to be capable of obtain a settling time for the differential output compatible with the photonic network. In conclusion, a faster and more energy efficient photonic neural network was developed than other more conventional neural networks.