Semiconductor Leading In-Memory Computing Technology
CIM (Compute-In-Memory) refers to a semiconductor approach where computation and storage are performed within the same region, rather than being separated as in the von Neumann architecture. A similar concept is PIM (Processing-In-Memory), but while CIM is a broader concept, PIM refers to specific architectures such as placing the processor near the memory to achieve this integration.
Ternell approaches CIM from the perspective of changing the memory cell structure itself, rather than simply adopting PIM. Leveraging ternary (three-valued) technology, Ternell offers a more compact and power-efficient ternary T-CIM (Ternary Compute-In-Memory) solution compared to conventional binary CIM. Additionally, through optimized analog-to-digital conversion and Multiply-Accumulate (MAC) operations, T-CIM supports optimal AI processing even in resource-constrained environments.
World’s first silicon implementation of T-CIM : Achieves a compact and efficient design by minimizing macro area overhead.
35% area reduction : Delivers excellent spatial efficiency through 1.5-bit ternary computing cells.
61% energy reduction : Dramatically reduces energy consumption compared to conventional binary systems through ternary input/weight-based zero-skip computation.
Hybrid bitline MAC : Improves energy efficiency via Ternell’s unique analog MAC operation combination, optimized for applications with limited battery capacity.
Double sampling ternary ADC : Provides twice the analog-to-digital conversion throughput compared to conventional binary ADCs, greatly enhancing AI inference performance.
Optimized signal margin : Ensures stable performance even in dense parallel operations through innovative analog CIM design.
