VALID IRREVERSIBLE RESOLUTION RATE: AN EVENT-LEVEL ENERGY METRIC FOR STABILIZED DECISIONS IN VLSI AND EDGE-AI HARDWARE

 Call For Papers..!!!

**Current Issue**

VALID IRREVERSIBLE RESOLUTION RATE: AN EVENT-LEVEL ENERGY METRIC FOR STABILIZED DECISIONS IN VLSI AND EDGE-AI HARDWARE

Martin Petrásek

ABSTRACT

Energy efficiency in VLSI and edge-AI hardware is commonly reported as OPS/W, TOPS/W, or energy per inference. These metrics quantify activity or completed tasks, but they do not require an explicit physical decision boundary, retention window, rejection rule, or stabilization ledger. This article develops the Valid Irreversible Resolution Rate (VIRR), an event-level metric that counts accepted stable irreversible outcomes per supplied joule. The framework separates full physical VIRR, task-level VIRR, and operationderived estimates to prevent overclaiming from incomplete public data. Reproducible demonstrations convert 27 MLPerf Tiny v1.3 power results into task-level outcomes/J, compute a VIRRpulse proxy from public CoO/Nb:SrTiO3 memristor data, and include a controlled full-pipeline Langevin validation with 

forward/reverse work and entropy-production ledgers. At a 5% update threshold, the resistance-based device-level VIRR estimate for 1.2 V pulses is 4.4× higher than for 1.4 V pulses.

KEYWORDS

event-level energy efficiency, irreversible decisions, information thermodynamics, VLSI, edge AI

For More Details Visit Here :

https://aircconline.com/vlsics/V17N4/17426vlsi01.pdf

Paper Submission URL : https://airccse.com/submissioncs/home.html