In a long discussion JDark convinced me that lost RAM is a software problem and everything is soldered correctly. I've smoked the intel manual, decompiled SPL - and found the reason - bootloader initializes timings only for the first 64-Mb memory bank! So when you access the second bank on stacked chips - you get just a garbage. Patching bootloader is a dangerous task, so i had to patch nk.exe. The second bug was in NK. As you know the last 1 mb of first 64Mb RAM is used for framebuffer, so we get a "hole" if we have 128 Mb RAM. Had to fix that too.
The test ROM is ready, but it is now too early to release it to public - the device hangs on exiting from sleep mode, but this could be fixed.
So to make 128Mb RAM working you'll need to solder chips, shorten CS1 as showed on the picture (other pins should be left as they are):
and flash a ROM when I'll publish it.
The ROM would be incompatible with G4 devices as long time ago I've broken something in my patches and don't have G4 to find what to fix. I need a person who has a G4 device and knows how to use KITL.
Now Universal has 112.44 Mb total RAM available to OS.
ASUS VivoTab RT TF600TG 64Gb 3G