#news from February: “#Micron Celebrates Opening of #India’s First Semiconductor Assembly and Test Facility” in #Gujarat. But not a fab: “converts advanced DRAM and NAND wafers from Micron’s global #manufacturing network into finished #memory and storage products”.
on 02026-04-10I didn’t realize Flash #hardware was now being shipped 48 layers of #memory cells deep. The two-dimensional shrink doubling time has now slowed to I think 27 months (Samsung was at 19nm at the beginning of 2014 and should be at 12nm at the beginning of 2017, an areal density increase of 2.5× in 36 months, putting the doubling time at 27 months) but they seem to be leapfrogging Moore’s Law by piling on layers, with the result that DIMM-sized SSDs now hold nearly a terabyte.
on 02016-12-07#paper on #memory #performance using “hybrid main memory” such as a DRAM cache layer over #Flash, as a stopgap until phase-change memory or something like that arrives.
on 02016-10-102009 #paper on improving #hardware #performance to get phase-change #memory to be very nearly as fast as DRAM, but nonvolatile. “A baseline PCM system is 1.6x slower and requires 2.2x more energy than a DRAM system. Buffer reorganizations reduce this delay and energy gap to 1.2x and 1.0x, using narrow rows to mitigate write energy and multiple rows to improve locality and write coalescing.” Presumably this is behind the #crosspoint stuff that Micron and Intel plan to launch next year? #persistent-memory
on 02016-10-10“Non-volatile storage: Implications of the Datacenter’s Shifting Center”, an ACM Queue #paper claiming that now #Flash “storage-class memories” of US$3000 to US$5000, such as NVDIMMs, provide durable #memory #hardware that is fast enough to saturate one or more fast CPUs, thus invalidating the assumptions that have underlain our memory hierarchy designs in computer systems ever since semiconductor RAM beat out ferrite cores in the early 1970s. They still only support about 100k IOPS though (or 400k for sequential or read-only workloads). and they still cost 25× as much as spinning rust #disks. (The massive Flash storage speed advantage is basically how #Fastly works.) It argues that now optimizing cost-performance requires us to be willing to let CPUs idle in order to keep the Flash beast fed, not vice versa, since the cost of the Flash rather than the CPU now drives the overall cost of the system. So, for example, maybe you should busywait on your Flash instead of using interrupts.
on 02016-01-06