“The increasing use of larger context windows and key-value caching in large language models requires more and higher-performance memory.”