INTEL, 2nm Chip Neural Processing

IBM just popped up with an exciting breakthrough in transistor technology—the world’s first 2nm chips. The news comes straight out of IBM’s semiconductor research facility Albany, New York, and means great things for the likes of the company’s partners,  such as Samsung, and not least its newest partner: Intel.

In fact, it might just prove immensely valuable in Intel’s race to catch up with rival  TSMC.

With 50 billion transistors in a chip the “size of a fingernail,” IBM claims the research chip’s “tiniest components are smaller than a strand of DNA.” With it, the company has apparently surpassed its previous 5nm chips in terms of size, and it’s even projected to have a 45% performance improvement over 7nm chips, as well as use 75% less energy (Tom’s Hardware).

IBM’s $3 billion ‘Seven Nanometers and Beyond’ research program is moving us out of the age of the FinFET designs of yesteryear, which were fine back when 22nm was the norm, but tend to throw up issues when you get down to the ridiculous scales of today’s chips. These new nano-sheet designs are part of the gate-all-around design that means we’ll no longer have to deal with FinFETs leaking current.

Not much has been divulged about the chip’s materials or transistor density yet, nor how the company plans to work around the inherent constraints such a small chip throws up in  terms of interconnect scaling (how they’ll all be connected up in situ).

What we do know is that, thanks to IBM’s recent partnership with Intel, the path is becoming ever clearer for Intel to match TSMC’s ever-smaller nodes.  Those which act as the building blocks for AMD’s CPUs and GPUs, including the best CPU for gaming right now, among many  other things.

In good time, too. IBM expects these 2nm nano-sheet processes to come tumbling out of its partner’s foundries in 2024, albeit late that year. Perhaps in time to meet TSMC’s 2nm chips head-on.

INTEL is collaborating with IBM on their 2nm process-point, nano-sheet methodology; which will undoubtedly lead to faster processors, and neural processing. 

VisualSim has existing processor-template models that only require parameter modification. Existing VisualSim model that one can alter setup/parameters as processor model:  
/Path_VisualSim/VS_AR/doc/Training_Material/Architecture/Processor/Basic_Processor_Model.xml

Web Reference: https://www.pcgamer.com/ibm-breakthrough-2nm-transistor-research-intel-cpu/