Research

Neuromorphic SNN + CNN Vision for High-speed Target Tracking

Predatory animals have specialized multi-pathway acquisition and processing of visual information for high-speed and accurate prey tracking. Convolutional NN with an optical camera and Spiking NN (SNN) with an event camera have the complementary advantage in spatio-temporal resolutions causing a trade-off between accuracy and processing speed. We design a fusion algorithm and demonstrate their seamless union for fast and error-tolerant target tracking of high-speed target tracking.

• Ashwin Lele, Yan Fang, Aqeel Anwar, Arijit Raychowdhury "Bio-Mimetic High-speed Target Localization with Fused Frame and Event Vision for Edge Application" Frontiers in Neuroscience - Neuromorphic Engineering (2022) (link)

Heterogenous RRAM + SRAM SoC for Multi-modal High-speed Tracking [video]

Hybrid processing using SNN and CNN needs modality-matched compute for optimizing the performance. We map the CNN on RRAM compute-in-memory and SNN on SRAM compute-near-memory for high throughput and power-gated low-power application.

• Ashwin  Lele*, Muya Chang*, Samuel D Spetalnick, Brian Crafton, Shota Konno, Zishen Wan, Ashwin Bhat, Win-San Khwa, YuDer Chih, Meng-Fan Chang, Arijit Raychowdhury "73.53 TOPS/W 14.74 TOPS Heterogeneous RRAM In-memory and SRAM Near-memory SoC for Hybrid Frame and Event-Based Target Tracking"  IEEE International Solid-State Circuits Conference (2023) (link)

Multi-modal Neuromorphic Frame+Event Vision for Optical Flow at Edge [video]

Event and frame pipelines for vision processing have advantages in speed and accuracy respectively. We design a leaky CNN filter for high-speed optical flow estimation on the event pipeline. It is periodically fused with the reliable frame-based flow for applications like the identification of fast-moving objects in the stationary background.      

• Ashwin Lele, Arijit Raychowdhury. "Fusing Frame and Event Vision for High-speed Optical Flow for Edge Application." IEEE International Symposium on Circuits and Systems  (2022) (link)

SNN-based Autonomous Learning of Gait for Hexapod Locomotion  [video]  

Insects walk using the coordinated movement of legs driven by a central pattern generator. We showed an autonomously learning spiking central pattern generator that learns a stable gait using rewards from balance and vision sensors like an animal would learn to walk with sensory feedback.

• Ashwin Lele, Yan Fang, Justin Ting, Arijit Raychowdhury "Learning to Walk: Bio-Mimetic Hexapod Locomotion via Reinforcement Based Spiking Central Pattern Generation" IEEE Journal on Emerging and Selected Topics in Circuits and Systems (2020) (link)

• Ashwin Lele, Yan Fang, Justin Ting, Arijit Raychowdhury. "Learning to Walk: Spike Based Reinforcement Learning for Hexapod Robot Central Pattern Generation" IEEE Artificial Intelligence Circuits and Systems (2020) (link)

• Justin Ting, Yan Fang, Ashwin Lele, Arijit Raychowdhury "Bio-inspired Gait Imitation of Hexapod Robot using Event-Based Vision Sensor and Spiking Neural Network" IEEE International Joint Conference on Neural Networks (2020) (link)

Closed Loop for Hexapod Navigation using Spike-only Processing

Spike-based locomotion is augmented with event-camera and SNN-based visual processing. This closes the loop from sensing to actuation for event-only compute fabric enabling alternate robotic hardware platforms.

• Ashwin Lele, Yan Fang, Justin Ting, and Arijit Raychowdhury. "An End-to-end Spiking Neural Network Platform for Edge Robotics: From Event-Cameras to Central Pattern Generation." IEEE Transactions on Cognitive and Developmental Systems (2021) (link)

Circuits Cost Reduction for On-chip Learning in SNN with RRAM Synapses [video]

On-chip spike-time dependent plasticity-based learning requires pulse shaping circuits in SNNs taking up the majority of circuit resources. A newly proposed NIPIN selector was shown to eliminate pulse shaping circuits with 35x area and 60x power reduction in neuron circuits in HSPICE and virtuoso

• Ashwin Lele, Anand Naik, Lakshya Bandhu, Bhaskar Das, Udayan Ganguly. "Circuit Cost Reduction for Online STDP Using NIPIN Selector as Timekeeping Device in RRAM Synapse" IEEE International Symposium on Circuits and Systems (2020) (link)

• Das, Bhaskar, Ashwin Lele, Pankaj Kumbhare, Jörg Schulze, and Udayan Ganguly. "PrMnO3-Based Memory and Si Time-Keeping Selector for Area and Energy Efficient Synapse." IEEE Electron Device Letters (2019) (link)

2T1R Bitcell Design For Fast Reading in RRAM-based Cache [video]

Highly resistive RRAM devices dissipate low power but result in longer read latency. This is because of the limited current available for charging the line capacitance. We propose and evaluate the 2T1R bitcell that separates the read and write current pathways for fast reading and low power application simultaneously.

• Ashwin Lele, Srivatsava Jandhyala, Saurabh Gangurde, Virendra Singh, Sreenivas Subramoney, Udayan Ganguly "Disrupting Low-write-energy vs. Fast-read Dilemma in RRAM to Enable L1 Instruction Cache"  International Symposium on VLSI Design and Test (2022) (accepted)

True Random Number Generator using differential OTP Devices 

Randomness in the hard breakdown of oxides can be used as a physical source of randomness for random number generation. We demonstrated the superiority of PECVD oxide-based OTP devices over gate oxides in MOSFETs and conceptualized and tested differential OTP-based unbiased random number generator  

• Pratiksha Malviya, Sunny Sadana, Ashwin Lele,  Kumar Priyadarshi et al. "A Differential OTP Memory based highly Unique and Reliable PUF at 180 nm technology node." Solid-State Electronics (2021) (link)

• Sadana, S., Ashwin Lele, S. Tsundus, P. Kumbhare, U. Ganguly. "A Highly Reliable and Unbiased PUF based on Differential OTP memory." IEEE Electron Device Letters (2018) (link)

• Udayan Ganguly, Sunny Sadana, Ashwin Lele. “System and method for generating random bit string in an Integrated Circuit” Indian Patent Application No. 201821010427, US Patent Application No. 16982153

• Ashwin Lele, S. Sadana, A. Singh, H. S. Jatana, U. Ganguly. "A simple PECVD SiO_2 OTP memory based PUF for 180nm node for IoT."  IEEE Device Research Conference (DRC), 2017 (link)

Musical Phrase Searching using Query Alignment in Hindustani Classical Music 

Matching the melodic query using the conventional DTW algorithm has a huge computation cost. We showed a more efficient Smith-Waterman local string alignment algorithm for query-based searching to reduce the complexity of the algorithm by an order in python

• Ashwin Lele*, S. Pinjani*, K. K. Ganguli, P. Rao. "Improved melodic sequence matching for query based searching in Indian classical music." Frontiers of Research on Speech and Music (2016) (link)

• K. K. Ganguli, Ashwin Lele*, S. Pinjani*, P. Rao, A. Srinivasamurthy, S. Gulati. "Melodic shape stylization for robust and efficient motif detection in hindustani vocal music". IEEE Twenty-third National Conference In Communications (2017) (link)

Ion Migration-based Reliability Analysis in Low-Cost Perovskite Solar Cells

Perovskites are prone to migrating ions which are expected to affect performance. We demonstrated charge accumulation based 55% potential degradation using Sentaurus TCAD

• Ashwin Lele*, S. Pinjani*, M. T. Patel, V. Nandal, S. Agarwal, P. R.Nair. "Effect of interface charges on the efficiency of perovskite based solar cells" Photovoltaic Specialists Conference (2016) (link)