arXiv subjects

Manan Suri

Publications and source records attributed to Manan Suri.

41 records · Page 3Linked to original sources

arXiv2018

Low-Power (1T1N) Skyrmionic Synapses for Spiking Neuromorphic Systems

In this work, we propose a `1-transistor 1-nanotrack' (1T1N) synapse based on movement of magnetic skyrmions using spin polarised current pulses. The proposed synaptic bit-cell has 4 terminals and fully decoupled spike transmission- and programming- paths. With careful tuning of programming parameters we ensure multi-level non-volatile conductance evolution in the proposed skyrmionic synapse. Through micromagnetic simulations, we studied in detail the impact of programming conditions (current density, pulse width) on synaptic performance parameters such as number of conductance levels and energy per transition. The programming parameters chosen used all further analysis gave rise to a synapse with 7 distinct conductance states and 1.2 fJ per conductance state transition event. Exploiting bidirectional conductance modulation the 1T1N synapse is able to undergo long-term potentiation (LTP) & depression (LTD) according to a simplified variant of biological spike timing dependent plasticity (STDP) rule. We present subthreshold CMOS spike generator circuit which when coupled with well known subthreshold integrator circuit, produces custom pre and post-neuronal spike shapes, responsible for implementing unsupervised learning with the proposed 1T1N synaptic bit-cell and consuming ~ 0.25 pJ/event. A spiking neural network (SNN) incorporating the characteristics of the 1T1N synapse was simulated for two seperate applications: pattern extraction from noisy video streams and MNIST classification.

physics.app-ph↗

arXiv2018

MASTISK

In this paper, we present MASTISK (MAchine-learning and Synaptic-plasticity Technology Integrated Simulation frameworK). MASTISK is an open-source versatile and flexible tool developed in MATLAB for design exploration of dedicated neuromorphic hardware using nanodevices and hybrid CMOS-nanodevice circuits. MASTISK has a hierarchical organization capturing details at the level of devices, circuits (i.e. neurons or activation functions, synapses or weights) and architectures (i.e. topology, learning-rules, algorithms). In the current version, MASTISK provides user-friendly interface for design and simulation of spiking neural networks (SNN) powered by spatio-temporal learning rules such as Spike-Timing Dependent Plasticity (STDP). Users may provide network definition as a simple input parameter file and the framework is capable of performing automated learning/inference simulations. Validation case-studies of the proposed open source simulator will be published in the proceedings of IJCNN 2018. The proposed framework offers new functionalities, compared to similar simulation tools in literature, such as: (i) arbitrary synaptic circuit modeling capability with both identical and non-identical stimuli, (ii) arbitrary spike modeling, and (iii) nanodevice based neuron emulation. The code of MASTISK is available on request at: https://gitlab.com/NVM IITD Research/MASTISK/wikis/home

cs.ET↗

arXiv2018

Design Exploration of Hybrid CMOS-OxRAM Deep Generative Architectures

Deep Learning and its applications have gained tremendous interest recently in both academia and industry. Restricted Boltzmann Machines (RBMs) offer a key methodology to implement deep learning paradigms. This paper presents a novel approach for realizing hybrid CMOS-OxRAM based deep generative models (DGM). In our proposed hybrid DGM architectures, HfOx based (filamentary-type switching) OxRAM devices are extensively used for realizing multiple computational and non-computational functions such as: (i) Synapses (weights), (ii) internal neuron-state storage, (iii) stochastic neuron activation and (iv) programmable signal normalization. To validate the proposed scheme we have simulated two different architectures: (i) Deep Belief Network (DBN) and (ii) Stacked Denoising Autoencoder for classification and reconstruction of hand-written digits from a reduced MNIST dataset of 6000 images. Contrastive-divergence (CD) specially optimized for OxRAM devices was used to drive the synaptic weight update mechanism of each layer in the network. Overall learning rule was based on greedy-layer wise learning with no back propagation which allows the network to be trained to a good pre-training stage. Performance of the simulated hybrid CMOS-RRAM DGM model matches closely with software based model for a 2-layers deep network. Top-3 test accuracy achieved by the DBN was 95.5%. MSE of the SDA network was 0.003, lower than software based approach. Endurance analysis of the simulated architectures show that for 200 epochs of training (single RBM layer), maximum switching events/per OxRAM device was ~ 7000 cycles.

cs.ET↗

arXiv2017

Dispenser Concept for Unmanned Aerial Vehicles (UAV, Drone, UAS)

System, design and methodology to load and dispense different articles from an autonomous aircraft are disclosed. In one embodiment, the design of a unique detachable dispenser for delivery of articles is described along with an intelligent methodology of loading and delivering the articles to and from the dispenser. Design of the dispenser, interaction of the dispenser with the flight control unit and ground control or base-station, and interaction of the base station with the sender or recipient of the article, are also described.

cs.CY↗

arXiv2017

Exploiting OxRAM Resistive Switching for Dynamic Range Improvement of CMOS Image Sensors

We present a unique application of OxRAM devices in CMOS Image Sensors (CIS) for dynamic range (DR) improvement. We propose a modified 3T-APS (Active Pixel Sensor) circuit that incorporates OxRAM in 1T-1R configuration. DR improvement is achieved by resistive compression of the pixel output signal through autonomous programming of OxRAM device resistance during exposure. We show that by carefully preconditioning the OxRAM resistance, pixel DR can be enhanced. Detailed impact of OxRAM SET-to-RESET and RESET-to-SET transitions on pixel DR is discussed. For experimental validation with specific OxRAM preprogrammed states, a 4 Kb 10 nm thick HfOx (1T-1R) matrix was fabricated and characterized. Best case, relative pixel DR improvement of ~ 50 dB was obtained for our design.

cs.ET↗