Nowadays, microprocessors and memories are made on different manufacturing lines. These lines are flexible 10 megabyte transistor microprocessors and 256 megabyte transistor DRAMs. Plants that manufacture these chips cost billions of dollars. One of the biggest challenges in computers today is the speed inequality between microprocessors and the memory. To deal with this challenge, processor designers devote a large part of the transistors and area of the chips to large SRAM caches.
Engineers predict that over the next decade, processors and memory will be put onto a single chip. This would narrow and altogether remove the processor-memory performance gap and it will have the following additional benefits: provide an perfect building-block for processing, lessen the costs of fabrication, and better utilize the number of transistors that can be placed on a single chip. Intelligent RAM or IRAM as they call it because most of the transistors on this fused chip will be committed to memory.
The goal of the IRAM is to show that a single chip with a plain and simple processor and a very high bandwidth local memory can be faster on memory problems. It will also be a much better match to real-time applications. Conventional machines will have separate chips for the processor, external cache, main memory, and networking, so an IRAM would also be smaller, use less power, and be less expensive.
A single gigabit of IRAM should have an internal memory bandwidth of almost 1,000 gigabits per second, more than a hundred times over the fastest computers today. Hence, the fastest programs will keep most memory accesses within a single IRAM, satisfying compact representations of code and data. .
IRAM's large improvement in memory system bandwidth has significant potential for helping re-configurable systems to reach their full performance. Re-configurable systems offer better performance by changing processing capabilities to the applications needs.