News

Systems for Sustainable Machine learning: With the easy availability of large accurate ML models and the rising interest in building AI applications, the overall ML workload, such as within a data-center, has shifted from model training to model inferencing. Therefore, optimising model inferencing has become crucial for green AI. Ongoing research includes improving latency and throughput of ML inference serving systems, and building cost-effective and sustainable ML workflows. 

 

http://abhilash-jindal.com/assets/popper

Faculty: Abhilash Jindal (https://abhilash-jindal.com/), 

Rijurekha Sen (https://www.cse.iitd.ac.in/~rijurekha/), 

Kaustubh Beedkar (https://web.iitd.ac.in/~kbeedkar/)

 

Embedded Systems and System-on-Chip Architecture: Several technological advances have converged to deliver sophisticated user experiences in the computing world: AI/ML applications, emerging hardware platforms, and novel fabrication and packaging technologies. This creates many opportunities for delivering higher system performance and compact solutions, as well as a large number of system design challenges. High performance needs to be delivered while respecting tight power and energy constraints. Large volumes of data need to be carefully orchestrated through future memory systems. Computation and data storage need to be subjected to complex partitioning, re-structuring, and mapping processes to ensure optimal utilisation. Processor architectures need to be augmented by application-specific hardware accelerators to address the throughput and latency requirements of modern applications. We address several research issues in this domain, including next-generation memory systems, AI/ML hardware, and 3D chip stacking.

 

https://marg.iitd.ac.in/

Lead faculty: Preeti Ranjan Panda (https://www.cse.iitd.ac.in/~panda/)

 

Biomaterial-based computing: Silicon microelectronics is still the workhorse of all commercial nanotechnology (1). Yet Nature uses proteins as tools to perform a dazzling variety or roles at the nanoscale. Taking inspiration from this, our research aims to fabricate computing devices from biological protein polymers. Presently, we are interested in polymers called microtubules, because they are both mechanically robust and highly abundant in eukaryotic cells. Although they play a variety of well researched structural roles inside the biological cell, the presence of a lattice-like arrangement of aromatic residues in microtubules also allows for unexpectedly efficient electronic energy migration (2-4). These properties, along with ease of chemical conjugation at room temperature raises the tantalizing prospect of microtubule photochemistry being used for computing applications (5, 6). Using previous work on DNA-based computing as a pathway for our research (7-9), we are interested in investigating how photochemical information in microtubules can be (A) read, (B) stored and (C) converted into different forms for pattern-based computing (for example, in cellular automata; see figure below). We will use approaches from fluorescence microscopy (to observe the microtubule lattice), ultrafast spectroscopy (to measure rates of electron transfer) to visualize microtubule-based automata. Furthermore, we are interested in using protein polymers for creating electronic devices for memory storage, light harvesting and energy transduction. We envision that our research on biomaterial-based computing will make significant contributions towards the thrust for ‘green computing’.

 

Faculty: Aarat Kalra (https://www.kalralab.org/)

 

ML-based technologies for industrial design

Lead faculty: Prof. Smruti R. Sarangi (http://www.cse.iitd.ac.in/~srsarangi)

 

Currently, industrial designs for 3D models and PCBs use slow manual tools that require a lot of designer expertise and have a steep learning curve. Our aim is to use fast AI-based algorithms (like Microsoft Copilot) to automatically generate engineering designs based on generative AI technologies. We are creating a PCB design tool that will use generative AI to compute the best PCB designs. It will be coupled with advanced GPU-based algorithms to simulate complex phenomena such that the gen-AI-driven optimization process converges quickly.

 

AR/VR for industry 4.0

Lead faculty: Prof. Smruti R. Sarangi (http://www.cse.iitd.ac.in/~srsarangi)

 

We still use conventional methods using mice, keyboards and 2D screens to design engineering elements such as PCBs, 3D designs and IIoT components. It is now possible to use modern AR, VR and haptics technologies to design these components in a novel and intuitive manner. We wish to design 3D in 3D. This project aims at creating a new kind of 3D design tool that uses the best in gen-AI and LLMs to create 3D structures.