The blog entry I wrote yesterday about the IHS iSuppli report on PCs and ARM processors set me on a thought path about PC design. (See “IHS iSuppli predicts that ARM CPUs will grab 25% of the PC market by the year 2015. Believe it?”) If the report is correct about ARM processor penetration into the PC market, then I think we can expect to see much faster development of the “1-chip” PC from both the x86 PC processor vendors and from the ARM processor chip vendors. Now such a PC wouldn’t really have just one chip. It would have a logic SoC with the CPU and a variety of IP subsystems including video and graphics controllers, memory- and storage-management controllers, and networking peripherals. Bulk memory in the form of DRAM and NAND Flash memory would still be off chip because commodity memory continues to be less expensive when packaged separately. This integration path isn’t very different from the one already occurring for mobile phone handsets and set-top boxes.
Letting my mind drift along that thought path reminded me of another recently announced report from Objective Analysis (“How PC NAND Will Undermine DRAM“) that discusses the mix of DRAM and NAND Flash memory in a PC. This in-depth report summarizes a series of benchmark tests and makes recommendations on finding the optimum mix of DRAM and NAND Flash for specific PC price and performance goals.
In a LinkedIn posting, Objective Analysis analyst Jim Handy wrote “We found that adding NAND to the PC (in paired storage or as a hybrid drive) adds more performance than adding DRAM in a dollar-for-dollar comparison. The gap between the two will widen over time, and is likely to wreak havoc on the DRAM market, while leaving HDDs relatively untouched. This was based on almost 300 benchmarks performed on a wide array of DRAM and NAND sizes in a typical PC. Kind of a surprising finding.”
Here’s a section from a graph in the report (generously on loan to the EDA360 Insider from Objective Analysis) that shows you the performance achieved with DRAM alone or with a mix of DRAM and NAND Flash using a fixed component cost for the two memory categories combined.
As the graph shows, a mix of NAND Flash and DRAM memory outperforms the amount of DRAM you can get for the same BOM (bill of materials) cost, up to a point where the black and red curves intersect. The reason for this performance difference is that the per-bit cost of NAND Flash memory is substantially lower than the per-bit cost for DRAM, and the price gap has been widening since 2004. You get much more NAND Flash memory per dollar of cost and that extra memory can hold more program code and more data, which in turn reduces the need to access the PC’s hard disk drive. Hence the performance improvement. (At the same time, NAND Flash memory is much slower than DRAM, so you cannot eliminate DRAM from a PC.)
According to the Objective Analysis report, things will get even more interesting by the year 2015—when ARM processors will supposedly ship in 23% of the new PCs sold that year.
By the way, these sorts of tradeoffs are exactly what System Realization teams should be making based on appropriate system-level simulations.
PS: Will NAND Flash entirely replace the hard disk drive as its price/bit continues to drop? That too is addressed in the Objective Analysis report.