Slashdot Mirror
24M gates + DRAM on a chip
dca writes "
It combines logic and DRAM on the same chip, allowing complete
high-powered systems on a single chip. Expected to be used first for
embedded applications like cell phones and palmtops. Based on their
copper wiring and trench DRAM; 24M gates; 0.15 micron process. It's
still in the lab today, so you'll be able to be the
first-on-your-block by summer. Makes me wonder how long the "new"
crop of Palms will last.
"
You'd expect the webmaster of a large company like IBM to know what their doing, but what's with the ?'s? Someone needs to yell at their webmaster...
- =^o.o^=
so if a microprocessor is a one-chip CPU, is this one-chip CPU and main store a micro-microprocessor? And do I now have worldwide copyright on the term? ;)
The IRAM (Intelligent RAM) project at Cal has been doing this for a while. They even have graduate class project using the shit (cs-274 I think is about prefetching on IRAM).
http://iram.cs.berkeley.edu
I'd hate to be the offtopic party gatescrasher... but there are two niggling issues 4 me:
i) Does IBM have have anything on the books to create mass storage, cheap flash/Dram/whatever chips to go with these advancements ? Are the memory manufacturers scared they'll go out of business with integrated micro-microprocessors (tm Anonymous Coward)?
Is this just a fad ?
ii) What good is the idyllic, fairly cheap, high performance chip in a PC system application if it takes a $1000 + performance SCSI/whatever HD to keep up ??????? Wheres the cutting edge absolutely needed 50Mb/s+ uncached/sustained transfer rates to meet the requirements of slow (or Microsoft's) platform? It's like having a Ferrari with model T Ford tyres.......Unimpressed with non-concurrent (realistic) advancements in technology.
We've all seen ideas that might rock the well established computer industry get squashed before they enter markets (PC's not just palmtops) where the masses actually need them.
Perhaps its the STANDARDS that are wrong!
Note that in the embedded market, integration of processor, main memory, and peripherals has been a big win for a long time. The PIC processor, for instance, costs less than $3 apiece and has CPU, memory, ROM, serial, and AD/DA integrated on chip. The primary advantage is reduced pincount -- there are only pins coming out of the chip for power, reset, AD/DA, and the serial connection (if any). A six-pin package is *much* easier and less expensive to integrate into your product (not to mention easier to use for the solder-slinging home hobbyist!) than a 300+ pin Ball Gate Array.
There are advantages to be had in the desktop computer market, too, for reducing pinout through memory integration. The latest crop of motherboards have a >32-bit address bus and a 128-bit main memory bus. That's at least 160 pins coming out of the processor just for that, and 160 traces on the motherboard, which is expensive in itself but more importantly it's hard to keep all those signals synchronized at 200MHz! The next step is a 256-bit external bus and very large address spaces, but I can't see a processor dedicating ~300 pins *just* to the memory subsystem! Nor can I see motherboards (especially SMP boards) handling the task and still be affordable.
On the other hand, if main memory could be integrated into the processor, you cease to have a pinout problem entirely. You can jack up the main memory bus to 256, 512, 1024 bits without complicating the processor pinout or the motherboard layout.
And once again, let me stress what a big deal this is to electronics geeks who like to build embedded systems in their workshops. We can buy a $40 Celeron which would give My Little Robot enough crunch power to perform rudimentary visual pattern recognition, but joe-random hobbyist probably can't build a memory subsystem around it. A motherboard costs another $120 or so, and adds mass and energy consumption to the design. What if My Little Robot is a remote-controlled model airplane being retrofitted for internal computer guidance? A motherboard and a power supply for it is completely unfeasible. Now if that hobbyist could plop down $50 for an all-in-one 20-pin processor with integrated everything (memory, clock, serial and parallel ports, ethernet (a la Motorola's PPC850)), then he's in business.
-
So be imaginative, and be audacious -- incredible possibilities are unfolding in front of us.
-- Guges
Seems like a good way to integrate large amounts of l1 and l2 cache on the chip. Dedicate about a million gates to 256K l1 cache, another 8 million to 8M L2 cache, and you're all set. With an offboard 32M L3 cache, we're ready to rock!
-- Too lazy to get a lower UID.
Yes, I know that IBM can fabricate things at 0.15 micron right now. They can probably do 0.13 for you if you ask nicely and pay large amounts of money. This is done using an electron synchrotron X-ray source and custom masks and processes. It can generate chips in research and proof-of-concept quantities, but is expensive and doesn't have the volume for actual production. Rest assured that many people, at IBM and elsewhere, are trying to figure out how to do volume production at those linewidths. However, don't expect it in the immediate future.
Also, IIRC Slashdot posted an article about an earlier version of this chip that you can have fabricated now, so this post shouldn't be a surprise to most people here.
I agree that I'm mainly nit-picking, but this whole E2k thing has me uptight about people talking about technology that isn't in production yet
A motherboard with sockets for many of these chips would make a cool desktop beowolf system.
Well, the first thing that popped into my feeble little brain was, this isn't really a new concept, it's just a faster version of an old concept. There have been microcontrollers around for years which included their own storage space onboard.
The article talks about it being an ASIC, not a general-purpose computing device, which must be why they're talking about it as targeted for embedded systems instead of the next generation of PC. Even so, I'd love to try one out on my little breadboard, hook it up to a keypad and see what it'll do!
Now what would be really cool is if they could throw in a Linux kernel in microcode (reminiscent of that microcontroller chip that has BASIC in microcode) so you could have "Linux on a chip"!
Have they cloned Bill already?
RUN FOR THE HILLS!
----- Documentation is worth it just to be able to answer all your mail with 'RTFM' - Alan Cox.