Allwinner H5 Board: DDR3 Compatibility Issues With Different Memory ICs

Hello everyone! It is my first experience with ARM and DDR.  I have designed the board with Allwinner H5 microprocessor and one DDR3 memory ic. Board works ok with only SAMSUNG K4B4G1646Q-HYKO but with any other everything is wrong. When I try to calibrate  memory I have awful "noise" in power lines (photo). I measure it in the out of power source and on capacitors near the memory's power pins result is the same. I try it with NT5CC256M16EP-EKT, K4B4G1646D-BCK0 and K4B4G1646D-BCK0.
I hope that someone can help me with this problem or just can explain why with one memory IC all is ok but with the other not.
Thanks. 

Hi Ivan-We're not ignoring you- it's just that you've stumped all of us! Please be patient-we'll try to investigate further and come up with an answer that at least "sounds" plausible.-Rick 

Hi,
If your power supply is truly dipping from 1300mV to 640mV, that will totally screw up the DDR3 and it's the wrong voltage.  DDR3 operates off a 1.5V supply, your scope trace shows 1.3V. Did you use a short ground lead on the probe to reduce inductive pickup?DDR2/3/4 use a Vref supply which is 1/2 Vdd, so for DDR3, VDD is 1.5V, Vref is 0.75V. The Vref voltage is used for all timing measurements. If Vdd drops below Vref, the DDR RAM will not work as it'll lose it's settings!I don't know your system setup, do you create Vref from Vdd using a potential divider or a nice tracking regulator like the TPS51116?For power supply dips at 118Hz/8ms intervals, he likely cause is something related to the periodic refresh of the device as refresh currents are some of the highest demands. If your power supply, as measured, is out of spec, it could cause the DDR3 device to 'gulp' more current, causing the supply blip.I'm assuming you are using a 4 or 6 layer PCB with dedicated power planes?What speed do you operate the DDR interface at?
Are you using On Die Termination (ODT)?There are many things that can affect the operation of DDR RAMs. Check the basics first then we can move on.

SKHynix provide some of the most approachable guides on DDR memory, for DDr3, take a look at https://www.skhynix.com/static/filedata/fileDownload.do?seq=211

HTH
Ian

To look at the power supply, I suggest a 500-ohm 10X resistive probe. Get a piece of RG-174 or RG-58, 50-ohm coaxial cable long enough to go from your board to your scope. Put a BNC connector on one end, to plug in to the 50-ohm 'scope input. Alternatively, you could cut a 50-ohm BNC jumper cable in half and make two probes. If your 'scope is 1-megohm only, put a 50-ohm feedthrough terminator between cable and 'scope. At the board end, connect the cable shield directly to ground, with zero lead length. Connect a 499-ohm resistor (using very short leads) from the center conductor to VDD. For 100X attenuation, change the resistor to 4.99K ohms. Keep track of the power dissipation in the resistor. These probes are not for high voltages. These probes have resistive inputs at high frequencies, so ring much less than standard 10X probes, which have capacitive inputs at high frequencies. The standard 10X probes were originally developed for probing vacuum-tube equipment at frequencies below 10 MHz, very different from what you are working with. Commercial resistive probes work into the GHz range.

Thanks for your answer. I hawe found some mistakes in my schematic wich cause supply blip. I conect clock with wrong polarity and have wrong connection is bank addres lines (dont know how my first memory IC work with that). Also I found that for normal operation of NT5CC256M16EP with my clock freqency 400 MHZ it needs to be powered with 1,5 V power  instead  of 1,35V.  After I fixed that, board began  working. But sometimes I still have drops out  in power lines, but it can apears because of i cut the board and connet wrong lines with shirt wires.  I use potential divider for Vref and 6 layer pcb where one layer is gnd plane and one for power planes other four layers I use for signals. I use On Die Termination.

Thanks I will try use it.