Work

Hybrid Chips for Training and Inference: A Unified Approach

Year: 2025

Type: article

Abstract: This article explores the evolution and integration of hybrid chip architectures designed to bridge the gap between training and inference in artificial intelligence systems. The article examines architectural innovations in shared memory and compute units, thermal management advancements, and applications in edge computing scenarios. It analyzes performance optimization through specialized software stacks, including compiler innovations, runtime systems, and deployment paradigms. The article also investigates future directions in reconfigurable computing elements, non-volatile memory integration, and domain-specific acceleration, highlighting the potential for improved efficiency and adaptability in next-generation AI hardware platforms.

Source: International Journal on Science and Technology

DOI: 10.71097/ijsat.v16.i1.2957

Author: Deepika Bhatia -

Language: English

Topic: Parallel Computing and Optimization Techniques

Subfield: Hardware and Architecture

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W4408943276

FWCI:

Cites:

Cited by:

Related to: 10