Direct Rambus Memory, Part 1 – The Basics

Pages: 1 2 3

This is the first installment in a two-part article about Direct Rambus memory technology. It includes a detailed description of the unique characteristics of Rambus and how it differs from SDRAM. In a future article I will compare the internal operation of SDRAM and Direct Rambus based PCs and examine the impact on processor performance.

What is Direct Rambus?

Direct Rambus is a set of technologies that includes high data transfer rate memory chips, an associated physical memory interface, and the signaling protocol and topology used to connect them together. It is the third generation of high-speed memory technology to be created by Rambus Inc. of Mountain View, California. Rambus does not manufacture Direct Rambus memory chips (DRDRAMs) or the application specific integrated circuits (ASICs) that interface to them but rather licenses the designs, patents, and technology to companies that do.

Several years ago Intel, the world’s largest and most powerful manufacturer of microprocessors and PC chipsets (the ASICs that hook the microprocessor up to memory and input/output devices), decreed that Direct Rambus would be adopted as the main memory technology of future personal computers. What’s noteworthy about this proclamation is 1) up to this point in time evolution in memory technology was gradual and pushed along by industry consensus, and 2) Direct Rambus represents a major change in virtually every aspect of memory systems – the memory chips themselves, the ways memory modules are connected to the chipset, the physical signaling protocol, data rates, clocking, and thermal management.

The Good, the Bad, and the Ugly

The good news about Direct Rambus is that it is faster than current memory technology, but the bad news about Direct Rambus is it is slower than current memory technology. How can something be both faster and slower? Because there are different aspects to memory speed.

Think of the data flow out of a memory system as being similar to the baggage carrousel at an airport. The delay between the start up of the carrousel and the appearance of the first suitcase is analogous to memory latency, while the rate at which subsequent items appear out of the chute is equivalent to memory bandwidth. Direct Rambus is strong on bandwidth. In fact, data can flow in or out of each DRDRAM data pin at up to 800 million bits per second (Mbps), a rate eight times faster than today’s memory standard bearer, the PC100 synchronous DRAM (SDRAM).

The bad aspect of Direct Rambus is latency. Rambus Inc. likes to say that its fastest DRDRAMs have the same 20 ns page access latency as the fastest PC100 SDRAMs. This is technically true when you consider the memory chip in isolation. But DRDRAM-based memory systems differ from SDRAM-based systems in several significant ways in how the chips are wired together and communicate with the memory controller ASIC which add tens of ns to the effective latency.

The ugly thing about Direct Rambus is the cost. As mentioned previously, Direct Rambus changes just about every aspect of the memory system and in each case it seems the changes add cost. The extra interface complexity and wider internal data paths cause a 128 Mbit DRDRAM to be about 20% larger in die size than a 128 Mbit SDRAM. Also, in a 0.22 um process only about 30% of working DRDRAMs are fast enough to be sold as top grade 800 Mbps parts. There are slower speed grades of DRDRAM (600 and 712 Mbps) but the latency of these parts are even worse which may make them unattractive for computer memory applications. The DRDRAM die is packaged in a micro-BGA or equivalent chips scale package which is significantly more expensive than the thin small outline package (TSOP) used for SDRAMs.

SDRAMs invariably end up attached to a small printed circuit board called a dual in-line memory module or DIMM, which plugs directly into a computer. The Direct Rambus equivalent to the DIMM is the Rambus in-line memory module or RIMM. Aside from the memory chips, a RIMM is generally more expensive than a DIMM because a RIMM must be manufactured with tighter tolerances for trace sizing, spacing, and thickness to ensure the near perfect transmission line environment needed for Direct Rambus’s very high data rates.

RIMMs also generally require a metallic heat spreader enclosure to avoid an excessive localized heating of any single memory device. Finally, the computer system motherboard into which RIMMs plug must have tightly controlled electrical characteristics that match RIMM circuit cards to avoid unwanted impedance mismatches and signal reflections. This can require extra signal layers and power planes, which along with the tighter manufacturing tolerances, results in a more expensive computer motherboard.

Pages:   1 2 3  Next »

Be the first to discuss this article!