News

After a long rebuild, the new site is live. Built with Astro, deployed via GitHub Pages, fully open-source at github.com/cyanmishra92/cyanmishra92.github.io.

The old sites.google.com/view/cyansubhramishra/ page is being deprecated — if you have it bookmarked, please update to https://cyanmishra92.github.io/.

Feedback welcome via the contact page.

Started as a Performance and Power Engineer on the SoC Architecture Performance team at Arm in San Diego. Working on pre-silicon performance and power analysis for next-generation heterogeneous SoCs targeting Edge AI, IoT, and automotive workloads.

Defended my Ph.D. thesis in Computer Science and Engineering at Penn State and officially graduated. Seven years in the Microsystems Design Lab, advised by Mahmut Taylan Kandemir and Jack Sampson.

Salient Store — our work on FPGA-accelerated computational storage for continuous-learning edge servers — was accepted at PACT 2025.

CORD — parallelizing query processing across multiple computational storage devices — was accepted at IPDPS 2025. Joint work with Wahid Uz Zaman, Saleh AlSaleh, Abutalib Aghayev, and Mahmut Taylan Kandemir.

NExUME — intermittency-aware DNN training that adapts dropout and quantization to harvested energy, with up to 22% accuracy gain at <5% compute overhead — was accepted at ICLR 2025.

Posted the Salient Store preprint to arXiv (2410.05435). FPGA-accelerated computational storage for continuous-learning edge servers — 6× throughput, 6.1× less data movement.

Posted the Synergistic and Efficient Edge-Host Communication for Energy-Harvesting WSNs preprint to arXiv (2408.14379).

Posted the NExUME preprint to arXiv (2408.13696). Revisiting DNN training for intermittently-powered energy-harvesting micro-computers.

Presented Usás — a sustainable continuous-learning framework for solar-powered edge servers — at HPCA 2024 in Edinburgh.

Presented our MICRO 2022 paper on pushing point cloud compression to the edge — a two-stage pipeline for geometry (parallel octree) and attribute compression with significant speedups and energy savings on edge platforms.

Presented Cocktail at NSDI 2022 in Renton, WA — multidimensional optimization for model serving in the cloud. Recording on YouTube.

Origin — on-device intelligence for human activity recognition using energy-harvesting wireless sensor networks — was nominated for Best Paper at DATE 2021. Joint work with Jack Sampson, Mahmut Taylan Kandemir, and Vijaykrishnan Narayanan.