-µarch implementation, RTL Design and integration of a function_present filter and 32 entry LUT for PCIe IP to route CPU PCIe configuration space access requests to backend software in case of a function miss in LUT-Implemented timeout mechanism to ensure that user software is not blocked out indefinitely from executing other configuration requests if backend software fails to acknowledge a request.-Implemented Messaged Signal Interrupt (MSI) feature to route errors to corresponding event collector -Definition and implementation of register and bit fields for various address spaces (PRB, IERB) and improved register reusability with other PCIe IPs-Identified and resolved input to output timing paths post synthesis to meet timing and area requirements

I worked as a project intern at the Council of Scientific and Industrial Research on the project titled "Identification of cognitive stress from EEG signals using Autoencoders". Recorded raw EEG signals from Emotiv brain sensor into a .CSV file. Using a 5th order butterworth filter, both EMG and power frequency are removed with lower cutoff frequency as 1 HZ and upper cutoff frequency as 40 HZ. This process is known as cleaning of EEG data. Then an autoencoder is trained and tested. An autoencoder is an unsupervised artificial neural network which learns efficient data coding. To train the autoencoder, some gaussian noise is to be added row wise, so that the autoencoder can be effectively trained to give clean data as the output removing the added noise. Thus the autoencoder learns by itself to automatically remove noise from data and to provide a clean EEG data. Then the autoencoder is tested by feeding an raw unprocessed EEG dataset and observe its efficiency. For identification of cognitive stress, we need to perform feature extraction and classification of the processed EEG data. This is done using a conventional neural network (CNN). The tested CNN can predict the state of mind of the user, but only limited to thinking words like COLLEGE, FRIENDS, LAB, CAT, DOG, and has an accuracy of 88%.