Estimating maximum current of DDR3L memory banks

[Steffen Koepf]

Hi,

i want to estimate the maximum current that can be consumed of a DDR3L
memory bank, consisting of 4 chips with 16 bits data bus width, so in
total 64 bit. I need this maximum current for the power supply design
(buck converter, 5V -> 1.35V).

I have read through the document TN-41-01: Calculating Memory System
Power for DDR3, it shows how to calculate the system power in a very
sophisticated manner. But there are a lot of variables i don't know
at the moment, like memory access patterns etc.

What about using the highest value of the datasheet, which seems to
be often I_DD5B, for example 250mA at a AS4C512M16D3L as the maximum
current, so 4 chips consume at max. 1A in total (DRAM Controller of
a SoC not included)?

And about the refresh, the datasheets note that a maximum of eight
refresh commands can be posted, does this mean, the DRAM executes
always only one refresh command at the same time, which refreshes
one eight of the DRAM, probably one bank, and the next refresh
command refreshes the next part of the dram?

Thanks in advance,

Steffen

[John Larkin]

I recently measured some supply currents on a microZed board. It has a
7020 ZYNQ and two DDR3 ram chips. The Zed schematic and such are
available online.

The board

http://www.highlandtechnology.com/DSS/P350DS.shtml

https://dl.dropboxusercontent.com/u/53724080/PCBs/P350_SN1.JPG

runs the Linux that comes with the zed. Bootup takes about 1 second.
Once it's running, some/most of the code is probably cached, so the
dram isn't working very hard.

+1 core supply 0.32 amps for 1 sec, then 0.6A

+1.5 ddr supply 0.23 amps for 1 sec, then 0.36 amps

I'll try to get an Ethernet connection, and then have some guys flail
it and play waveforms and things, try to exercize the dram more.

The TLV62130 switchers on the uZed are rated "up to three amps"

Let me know if you measure anything interesting.

I'm designing a new box that will use the 7020 and one DDR3 chip. "500
megabytes should be enough for anyone." My little switcher chips are
good for 2 amps, maybe 2.5, so I should be OK.



--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

[John Larkin]

This is really stupid:

I'm planning to use a bunch of cute little TPS54302 switchers to make
my supply voltages from a +24 source. I'll switch to +5, +3.3, +1.5,
and +1.0. A separate Cuk will make -6, and a lot of linears will make
+22, +12, +4, +3, +2.5, -2.5, -3, -5, and maybe more.

The question was whether it's better to switch directly from 24 to
1.0, or to switch 24 to 5 and then 5 to 1.0. At 400 KHz, and 1/24 duty
cycle, the switcher will be on for about 100 ns, going from +24 to
1.0.

So I hacked the eval board down to 1 volt output. But I can't get the
loop compensation right: the output keeps oscillating. Envision an
hour or two of fiddling. Finally, at the end of the day on Friday, I
notice that my 1 ohm test load is a giant wirewound resistor. So I
measure its inductance. 20 uH.

OK, fresh start on Monday.

--

John Larkin Highland Technology, Inc

lunatic fringe electronics

[krw]

On Fri, 07 Oct 2016 19:31:10 -0700, John Larkin

I always switch down in two stages, even from 12-16V. Lately I've
been using 3.8V as my primary and either bucking or boosting from
there. 3.8V because I could see intermittent dips to 4.5V and it's
easier to just make the thing run that low than adding capacitance to
hold the primary rail up. I try to synchronize all of the switchers,
and generally run them at around 2MHz (so get really concerned about
duty cycle - at both ends). I do use some cheap boosts that run at
3.5MHz, or so. I don't worry about them beating against the others
much. The one thing to watch out for is coupling the primary and
secondary switchers too tightly and make sure there is plenty of input
capacitance on the secondaries.

>So I hacked the eval board down to 1 volt output. But I can't get the
>loop compensation right: the output keeps oscillating. Envision an
>hour or two of fiddling. Finally, at the end of the day on Friday, I
>notice that my 1 ohm test load is a giant wirewound resistor. So I
>measure its inductance. 20 uH.

Oops.

[John Larkin]

The TPS54302 is a sot-23/6 that's rated for 3 amps, realistically
maybe 2.5. It runs at 400 KHz spread-spectrum, so I expect it to be
very nice as regards radiated EMI.

We have had the debate, synchronize or not? I vote not, to keep the
EMI things from summing up. The TPS can't be synchronized, so that's
not an issue here.

[krw]

On Fri, 07 Oct 2016 21:29:21 -0700, John Larkin

With a good layout, EMI shouldn't be a problem. I find it easier at
high frequencies because the parts are so much smaller. Our EMI
requirements are much tighter than class-B. The only time we have
problems with power supplies is if someone botches the design badly
(or some dumbass tries a 2-layer board).

Spread-spectrum is a no-no in our organization. Too much chance of
beating causing beating in the radios or even audio. Synchronizing
the switchers does several other nice things. You can set the
frequency where you don't care as much about EMI and it's easier to
track EMI down to the offending power supply, should there be an
issue.

TI makes tons that are, though not in SOT23-6, obviously. They have a
few new packages that further reduce EMI, as well.