447 TB/cm² at zero retention energy – atomic-scale memory on fluorographane
A new memory architecture using single-layer fluorographane can achieve 447 terabytes per square centimeter with zero retention energy. This innovation aims to address the widening gap between processor throughput and memory bandwidth, exacerbated by increased AI demand and a NAND flash supply crisis.
zenodo.org
1 min
4/11/2026
David Patterson: Challenges and Research Directions for LLM Inference Hardware
Large Language Model (LLM) inference faces significant challenges primarily related to memory and interconnect issues rather than compute power. The autoregressive Decode phase of Transformer models distinguishes LLM inference from training, complicating the process.
arxiv.org
2 min
1/25/2026
447 TB/cm² at zero retention energy – atomic-scale memory on fluorographane
A new memory architecture using single-layer fluorographane can achieve 447 terabytes per square centimeter with zero retention energy. This innovation aims to address the widening gap between processor throughput and memory bandwidth, exacerbated by increased AI demand and a NAND flash supply crisis.
zenodo.org
1 min
4/11/2026
David Patterson: Challenges and Research Directions for LLM Inference Hardware
Large Language Model (LLM) inference faces significant challenges primarily related to memory and interconnect issues rather than compute power. The autoregressive Decode phase of Transformer models distinguishes LLM inference from training, complicating the process.
arxiv.org
2 min
1/25/2026
447 TB/cm² at zero retention energy – atomic-scale memory on fluorographane
A new memory architecture using single-layer fluorographane can achieve 447 terabytes per square centimeter with zero retention energy. This innovation aims to address the widening gap between processor throughput and memory bandwidth, exacerbated by increased AI demand and a NAND flash supply crisis.
zenodo.org
1 min
4/11/2026
David Patterson: Challenges and Research Directions for LLM Inference Hardware
Large Language Model (LLM) inference faces significant challenges primarily related to memory and interconnect issues rather than compute power. The autoregressive Decode phase of Transformer models distinguishes LLM inference from training, complicating the process.
arxiv.org
2 min
1/25/2026
No more articles to load