Trends in semiconductor memories

Abstract

Despite their great market success, DRAMs have not kept pace with microprocessor improvements, so researchers are looking to advanced high-speed DRAM and merged DRAM/logic technologies to increase memory system performance. . Yasunao Katayama is an advisory researcher at IBM Research, Tokyo Research Laboratory, IBM Japan, Ltd. He has been engaged in research on positron physics, quantum devices, highspeed memory design, merged DRAM/ logic, and memory architecture. Katayama received BS and MS degrees in physics from Tokyo University and MS and PhD degrees in electrical engineering from

Optical Interconnect Opportunities for Future Server Memory Systems

“…Roughly speaking, the system design point (memory bandwidth and capacity vs. CPU performance) can be balanced through empirical rules such as Amdahl's rule of thumb and the ratio between memory bandwidth and capacity defines the fill frequency. If this trend continues, we will eventually put memory and CPUs on the same chip [3].…”

Architectures for real-time volume rendering

“…In the last 30 years the speed of microprocessors has increased 1000-fold whereas the speed of DRAM has only increased by a factor of 20 [Sak97]. As shown in Figure 1, recent architectural improvements to DRAM modules have increased their sequential access bandwidth [Kat97]. However, their random access performance remains low and approximately constant across different DRAM technologies.…”

Influence of Input/Output Operations on Processor Performance

“…Furthermore, new system buses [8][9][10] have been developed, with a higher bandwidth in order to improve data transfer speeds among computer devices (processor, memory, and I/O peripherals). Memories 11 and caches 2,4,12 have evolved too. Finally, technological advances will allow to increment the complexity of the chip and its working frequency.…”