Akshat Ramachandran

I am a Ph.D student in Electrical and Computer Engineering (ECE) at the Synergy Lab in the Georgia Institute of Technology (Georgia Tech) advised by Prof. Tushar Krishna.

Google Scholar  /  CV  /  LinkedIn  /  GitHub  /  Email

Community Service: In an effort to give back to the community, I conduct a weekly mentorship/guidance session for undergraduate/graduate students. Please read the description in the form and if you feel this could benefit you, please fill out the form. This effort is motivated by the lack of structured guidance in my early years of undergrad. (Note: On travel currently, please expect delays in response.)

profile photo
Research Interests

The central focus of my research is in the design and development of efficient and high performance algorithms, architectures and systems for accelerating emergent deep learning applications (computer vision and NLP). My research breaks down traditional barriers existing between different computing elements and adopts an interdisciplinary approach spanning the entire computing stack. To realise my vision of developing the next generation of computing systems, my interests and technical work spans a wide gamut and lies at the intersection of computer architecture, VLSI, computer arithmetic and deep learning.

  • Architecture:
    • Domain-Specific Architecture for Computer Vision: [DAC'24] [DSD'22]
    • AI for Hardware Design Space Exploration: [DATE'25]
News

Publications (Since 2024)       (*: Equal Contributions)
AIRCHITECT v2: Learning the Hardware Accelerator Design Space through Unified Representations
Akshat Ramachandran*, Jamin Seo*, Yu-Yuan Chang, Anirudh Itagi, Tushar Krishna
*Equal Contribution
Design Automation and Test in Europe (DATE), 2025
Paper

Design space exploration (DSE) plays a crucial role in enabling custom hardware architectures, particularly for emerging applications like AI, where optimized and specialized designs are essential. With the growing complexity of deep neural networks (DNNs) and the introduction of advanced large language models (LLMs), the design space for DNN accelerators is expanding at an exponential rate. Additionally, this space is highly non-uniform and non-convex, making it increasingly difficult to navigate and optimize. Traditional DSE techniques rely on search-based methods, which involve iterative sampling of the design space to find the optimal solution. However, this process is both time-consuming and often fails to converge to the global optima for such design spaces. Recently, AIRCHITECT V1, the first attempt to address the limitations of search-based techniques, transformed DSE into a constant-time classification problem using recommendation networks. However, AIRCHITECT V1 lacked generalizability and had poor performance on complex design spaces. In this work, we propose AIRCHITECT V2, a more accurate and generalizable learning-based DSE technique applicable to large-scale design spaces that overcomes the shortcomings of earlier approaches. Specifically, we devise an encoder-decoder transformer model that (a) encodes the complex design space into a uniform intermediate representation using contrastive learning and (b) leverages a novel unified representation blending the advantages of classification and regression to effectively explore the large DSE space without sacrificing accuracy. Experimental results evaluated on 105 real DNN workloads demonstrate that, on average, AIRCHITECT V2 outperforms existing techniques by 15% in identifying optimal design points. Furthermore, to demonstrate the generalizability of our method, we evaluate performance on unseen model workloads and attain a 1.7x improvement in inference latency on the identified hardware architecture.

MicroScopiQ: Accelerating Foundational Models through Outlier-Aware Microscaling Quantization
Akshat Ramachandran, Souvik Kundu, Tushar Krishna
Under Review
Paper

Quantization of foundational models (FMs) is significantly more challenging than traditional DNNs due to the emergence of large magnitude features called outliers. Existing outlier-aware algorithm/architecture co-design techniques either use mixed-precision, retaining outliers at high precision but compromise hardware efficiency, or quantize inliers and outliers at the same precision, improving hardware efficiency at the cost of accuracy. To address this mutual exclusivity, in this paper, we propose MicroScopiQ, a novel co-design technique that leverages pruning to complement outlier-aware quantization. MicroScopiQ retains outliers at higher precision while pruning a certain fraction of least important weights to distribute the additional outlier bits; ensuring high accuracy, aligned memory and hardware efficiency. We design a high-throughput, low overhead accelerator architecture composed of simple multi-precision INT processing elements and a novel network-on-chip called ReCoN that efficiently abstracts the complexity of supporting high-precision outliers. Additionally, unlike existing alternatives, MicroScopiQ does not assume any locality of outlier weights, enabling applicability to a broad range of FMs. Extensive experiments across various quantization settings show that MicroScopiQ achieves SoTA quantization performance while simultaneously improving inference performance by 3x and reducing energy by 2x over existing alternatives.

CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs
Akshat Ramachandran, Souvik Kundu, Tushar Krishna
European Conference on Computer Vision (ECCV), 2024
Paper

We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacting quantization accuracy. CLAMP-ViT employs a two-stage approach, cyclically adapting between data generation and model quantization. Specifically, we incorporate a patch-level contrastive learning scheme to generate richer, semantically meaningful data. Furthermore, we leverage contrastive learning in layer-wise evolutionary search for fixed- and mixed-precision quantization to identify optimal quantization parameters while mitigating the effects of a non-smooth loss landscape. Extensive evaluations across various vision tasks demonstrate the superiority of CLAMP-ViT, with performance improvements of up to 3% in top-1 accuracy for classification, 0.6 mAP for object detection, and 1.5 mIoU for segmentation at similar or better compression ratio over existing alternatives.

Algorithm-Hardware Co-Design of Distribution-Aware Logarithmic-Posit Encodings for Efficient DNN Inference
Akshat Ramachandran, Zishen Wan, Geonhwa Jeong, John Gustafson, Tushar Krishna
ACM/IEEE Design Automation Conference (DAC), 2024
Paper

We present Logarithmic Posits (LP), a novel adaptive data type for Deep Neural Network (DNN) quantization, offering significant efficiency and accuracy improvements. Our approach, LP Quantization (LPQ), optimizes DNN parameters using a genetic algorithm, closely preserving model accuracy with a specialized objective. The resulting LP accelerator architecture (LPA) doubles performance per area and improves energy efficiency by 2.2x over traditional quantization methods, with minimal accuracy loss.

Honors and Awards
  • [2024] Second position in ACM Student Research Competition at MICRO 2024.
  • [2024] Recepient of MICRO 2024 Student Travel Grant.
  • [2024] Selected as DAC Young Fellow Class of 2024.
  • [2022] Received the Samsung Spot Award for outstanding contribution towards design and development of AR/VR algorithms and acceleration on Samsung Galaxy devices.
  • [2022]Won the Best Undergraduate Thesis Award in the EE dept. for the senior year thesis on ”Next Generation Architecture for Computer Vision”.
  • Experience

    Intel Labs, Santa Clara, CA, USA
    AI Hardware Research Intern • Aug 2024 to Dec 2024

    NVIDIA, Santa Clara, CA, USA
    Architecture Energy Modelling Intern • May 2024 to Aug 2024

    Lemurian Labs, Menlo Park, CA, USA
    Hardware Intern (Remote) • Jun 2023 to Aug 2023

    Samsung Research, Suwon, South Korea & Bangalore, India
    Visual Intelligence Research Engineer • Aug 2022 to Jul 2023


    Academic Service

    • Conference Reviewer: CoNGA'24@SC Asia 24.
    Activities and Picture Gallery

    • Sports: I enjoy basketball, swimming, tennis and table tennis.
    • Music: I am a trained veena and violin player. [YouTube]
    Synergy@CoCoSyS

    CoCoSyS Members

    Image 2

    Deep Dive Session @ CoCoSyS

    Image 3

    Lightning Talk @CoCoSyS

    Image 4

    Synergy Lab@CoCoSyS

    Image 5

    Hiking @ Long Beach

    Image 6

    Whale Watching @ Cape Cod

    Image 7

    Veena

    Image 8

    State-Level Basketball Tournament


    Last Update: Nov. 2024

    Website Credits