17 Aug
2006
17 Aug
'06
6:16 p.m.
Dave McGuire wrote:
On Aug 17, 2006, at 7:39 PM, Tony Duell wrote:
You need to stagger refresh in big arrays. Just like gating
RAS/CAS to only those banks that *need* to see them for *this*
particular cycle -- for exactly this reason. The problem
is present even on small arrays.
There were a lot of myths about how critical the
layout was with DRAM
(and Rainbow printed some of them). Yes, you do need to take care. It is
a high-speed circuit, you should try to keep traces the same length,
decoupling is _essential_ as is a low-impedance ground track. But to be
honest, making an SRAM board that runs at the same speed is no easier.
From 1986 to early 1988 I worked on the Navier-Stokes Supercomputer
Project at Princeton University. Each node of that machine had four 4MW
memory planes (36-bit word) built from 41256 chips; 576 chips per memory
plane, handled by a pair of Intel 8207 DRAM controllers. We had really
nasty problems with the refresh cycles creating tons of noise on the Vcc bus. Man that was a nightmare; it took weeks to get
it cleaned up. If
I recall correctly we wound up rebuilding the boards with a bypass
capacitor for every DRAM chip.
<grin> I worked on a "600 pin tester" (i.e. be able to
stimulate and monitor the states of 600 different signals
on the unit-under-test) in the late 70's. The "stimulus
memory" was two "doors" (large -- ~18" x ~60" -- wirewrapped
panels on hinges fed with *thick* copper bars for power
distribution) full of bipolar & ECL memory devices.
I.e. *hundreds* of amperes.
Due to the mix and match of logic families (CMOS/TTL/ECL),
you had lots of odd supply voltages on the boards (-5.2,
-1.2, +5, +12, etc.) -- in addition to the other supplies
for the "pin drivers" (the UUT is driven with signal
levels that are programmed from a variable range of output
voltages).
Needless to say, *lots* of decoupling capacitors *everywhere*.
When the boards came in from the wirewrap house, there was a
short between two of the supplies (or, perhaps, a supply
and ground). Trying to figure out *where* the short could be
in a panel of that size COVERED with teflon wires was HUGELY
intimidating!
Long story short: the bypass caps were the wrong voltage.
So, as soon as power was applied, the HUGE power supplies would
very happily fry each cap to a dead short.
Happy ending: wire wrap vendor had installed the caps so
wire wrap vendor had to do the rework! ;-)