I can smile now, but I wasn't very happy last night at 2AM crawling around the floor of my basement looking for it!
I bought one of the cheapest trinocular metallurgical microscopes that I could find, an Omano OMM200 from microscope.com. It came with 10x and 16x eyepieces and 4x, 10x, 40x and 63x objectives. I swapped out the 63x with a more useful 20x. I generally use the 10x for full die shots and sometimes the 20x for a close up of the ROM array. I also bought a camera adapter, but I've since switched to using a 50mm lens digiscoping the 10x eyepiece. I 3d-printed a threaded cone to hold the camera in alignment with the eyepiece.
I built a simple XY table using a couple of linear slides and 2 linear actuator stepper motors. I put another stepper motor on the focus knob, but I recently added a kinematic mount to level the die manually. That helps with focus, since the die is parallel with the film plane.
The latest die I'm working on, from Entex Color Football,is a TMS1400 that measures 5.85mm x 4.82mm. I shot 208 pics of the full die (16x13) using the 10x objective, which resulted in a 19700x16200 pixel image, about 3.4 pixels per micron. To make it easier to work with, I scaled it down about 2.5 times.
I still haven't had luck running any automated panorama scripts, so I manually align the pics using a program I wrote, then use Microsoft ICE to create the composite, which I clean up in GIMP. I've got another program that reads the ROM array to a binary file, then I manually verify the bits by drawing them over a picture of the ROM array.
Blowtorch to ROM file is a couple of hours per chip at best.
