The Next-Generation Championship in Branch Prediction (CBP-NG)  1st Place

Award winners

Assistant Professor KOIZUMI Toru (Computer Science Group)
Professor TSUMURA Tomoaki (Computer Science Group)
Mr.MIZUNO Masanari (Master's Course, Program of Networks)
Mr.NISHIDA Soma (Master's Course, Program of Networks)

Related Website of the winner

Tsumura and Koizumi Lab.(Link to Japanese website)

Related Website

The Next-Generation Championship in Branch Prediction（CBP-NG）

Outline

At CBP-NG, held in conjunction with ISCA2026, a top-tier conference in computer architecture, a team led by Toru Koizumi of this university won first place. The same team also won CBP2025 last year, making this a second consecutive victory.

CBP (Championship Branch Prediction) is an academic competition on branch-prediction algorithms, a technique of growing importance in modern high-performance CPUs. TAGE, a past CBP winner, and its derivatives are used in commercial processors, making CBP a competition that shapes the field.
CBP-NG was held as a "next-generation" edition with renewed criteria. Previously, prediction accuracy under a fixed storage budget was essentially the sole criterion; CBP-NG also evaluates the energy and latency of prediction, since a predictor with high energy or latency costs cannot be adopted in a real processor, however accurate it may be.

The team proposed MORSL, a branch predictor that minimizes the circuitry activated on each prediction while preserving accuracy. Mechanisms such as accessing tables only when necessary greatly reduce wasted energy and latency, and MORSL scored highest among the teams worldwide, retaining first place under the renewed criteria: back-to-back victories.