The computer chip heads by the side of IBM's seek and Microelectronics divisions maintain had a taken week, viewing inedible their fresh remembrance and computing toys by the side of the IEEE's International Electron policy discussion in the sphere of Washington, D.C., carry on week. A little of the computer chip equipment with the aim of immense Blue was viewing inedible to its peers is far dazed into the opportunity, but it may well conclusion up in the sphere of systems quicker than you might think.
The same as you old AS/400 shops are well aware, IBM used to progress to its own remembrance chips and CD-ROM drives in favor of its systems and with the aim of remembrance was designed and fabbed in the sphere of the same Rochester, Minnesota, labs someplace the AS/400 and its forebears were all born. The magnetoresistive CD-ROM heads and super-dense CD-ROM platters came dazed of Rochester in the AS/400 C run generation, and so did dense DRAM key remembrance back in the sphere of the D run generation. (I used to maintain single of these chips, which I think was single of the at the outset 1Mbit chips continually, on my desk, but I maintain rebuff suspicion someplace it went). IBM in addition figured dazed how to stack up cheaper DRAM chips in the sphere of twos and fours with AS/400 and iSeries tackle back by the side of the dawn of occasion so it may well progress to super-dense remembrance cards dazed of relatively low-cost DRAM chips. IBM still knows a point or else two more or less remembrance, the same as evidenced by its L1 and L2 caches in the sphere of its mainframe and Power Systems processors and the embedded DRAM (eDRAM) with the aim of it position into the Power7 and z10 processors carry on time.
Perhaps other importantly in favor of a fresh remembrance equipment called Hybrid remembrance Cube (HMC), which is a fresh kind of 3D remembrance stacking formed by Samsung Electronics and Micron equipment, IBM has lots of expertise in the sphere of well ahead computer chip making and has approach up with a little breakthroughs with the aim of strength of character allow in favor of HMC memory--not to be present baffled with the Hardware Management Console in favor of Power Systems servers--to approach to marketplace.
Here's the suspicion behind 3D remembrance. If you stack up lots of plain-old DDR3 remembrance chips, link them as one vertically with a substrate of wires, snap a crossbar interconnect against folks wires to link the chips all as one, and interface them to a motherboard socket, after that you can run remembrance in the sphere of a other dense and analogous form than you can execute with standard 2D DDR3 remembrance modules and sockets now. By allowing remembrance access to run in the sphere of analogous across a ample car front-ended by with the aim of crossbar, you can run it slower and as a consequence generate a lesser amount of cook and still maintain a tremendous amount of bandwidth concerning the remembrance blocks and the processors with the aim of need the data on single region and the I/O systems with the aim of feed data into remembrance from the other region.
Under a deal with the aim of IBM cut out with Micron carry on week, the remembrance maker based in the sphere of Boise, Idaho strength of character be present licensing a little intellectual property with the aim of immense Blue cooked up to manufacture the linking wires in the sphere of the HCM lump, called Through Silicon Vias or else TSVs.IBM and Micron are not adage exactly what did you say? They maintain figured dazed, but they are made of copper and integrate well with the computer chip baking processes with the aim of Micron has in the sphere of its Idaho fabs. The same as part of the deal, IBM is in addition making the HMC logic circuits with the aim of realize the HMC crossbar and pinout to the motherboard, which strength of character be present implemented in the sphere of immense Blue's copper/high-k metal gate computer chip processes, which are implemented in the sphere of 32 nanometers. IBM strength of character progress to these logic circuits and push them to Micron, but it is Micron with the aim of strength of character be present making the DRAM and stacking it up with the TSVs and after that hooking the remembrance logic to the lump.
With HMC remembrance, a smidgen of data can be present transferred from remembrance to the CPU interconnect with more or less 70 percent a lesser amount of energy than with standard 2D DDR3 remembrance and an on-chip controller, and a regard office of key remembrance strength of character take up more or less one-tenth the hole. And for the reason that of the analogous nature of the interconnect concerning the HMC and the logic board, prototypes maintain been able to send 128 GB/sec of bandwidth dazed of the HMC lump, which compares quite auspiciously to the 12.8 GB/sec you can acquire dazed of a DDR3 remembrance stick running by the side of 1.33 GHz using a 2D configuration.
Micron expects to maintain HMC remembrance arrange in favor of marketplace by the succeeding partly of 2013, and head waiter makers are already lining up to see to it that how it might fit into their logic designs. Such super-dense, low-power remembrance modules are unquestionably crucial in favor of opportunity exascale supercomputers, but it remains to be present seen whilst it might appear in the sphere of broad-spectrum target servers like folks running net servers and databases.
IBM in addition thought carry on week by the side of the IEDM event with the aim of it has approach a smidgen closer to commercializing one more remembrance equipment, called racetrack remembrance.
I told you all more or less racetrack remembrance back in the sphere of January, and it is a actually funky equipment with the aim of promises remembrance densities with the aim of are other than two guidelines of magnitude better than CD-ROM areal density while by the side of the same occasion being other energy efficient and a lesser amount of expensive to progress to than explode remembrance. Disks move yet to be in excess of media to read data while tape moves media in excess of a stationary president to read it. Racetrack remembrance encodes data on zillions of nano-scale wires and uses spintronics to move the data up and down a sphere of wire the same as if it were a succinct part of a set of tape. But in the sphere of this legal action, the spintronics effect lone moves the data, not the wire, and it can execute it by the side of hundreds of miles an hour in excess of folks tiny wires. If you position a allotment of loops as one and a akin analogous access method and crossbar, you can progress to a very high-pitched bandwidth, low latency remembrance device.
With carry on week's notice by the side of IEDM, IBM was able to integrate the nanowires and the read and write down heads in favor of the loops, the same as well the same as the circuits with the aim of encode the data magnetically on the wire and depletion quantum-mechanical spin to move with the aim of data up and down the wire the same as if it were on a virtual part of a set of tape. Other importantly, IBM was able to fabricate the racetrack remembrance device, consisting of 256 racetracks, using standard CMOS processes on 200 millimeter wafers. The racetrack wires were more or less 150 nanometers ample, 20 nanometers thick, and 10 micrometers prolonged, which is pretty hefty by new circuit values. And as a consequence far, IBM is lone able to move single smidgen of data up and down the loops and needs to be present able to import lots of bits in favor of racetrack remembrance to be present practicable.
You'll see to it that HMC remembrance in the sphere of servers well or else racetrack remembrance.
Thinking way dazed into the opportunity, IBM's techies are in performance around with carbon nanotube transistors and thought carry on week by the side of IEDM with the aim of they were able to progress to a carbon nanotube transistor with channel lengths with the aim of were less significant than 10 nanometers, which is better than silicon-based computer chip fabbing equipment can execute. Heaven lone knows whilst we strength of character switch computers from silicon to carbon-based circuits.
没有评论:
发表评论