You'll find the details in the PPC Journal circa 1978/1979. It was a very simple device: a small circuit board was plugged into the power jack on the back of the '67 between the spring plate and the outer two pins. It had a section of the copper removed down its middle and a short cable attached. In the box was a potentiometer (actually, just a variable resistor) and a normally-closed pushbutton in parallel with it.
When the board was inserted, power from the 67's battery had to travel through the box to the rest of its circuitry. By pushing the button, the resistance was added to the battery. IF the resistance was close to zero, nothing happened. If it was large, the '67 saw its power drop to where it turned off while the button was pushed. But if the resistance was just right, the '67 would stop operating *but* its power-on reset circuit wouldn't be triggered.
The result was the 67's registers, being dynamic circuits, would forget their contents but not get reinitialized. This allowed access to making all sorts of non-normalized numbers, exploring around the CPU's address space, etc. Again, check the PPC Journals for the details.
I used this trick to create arbitrary bit patterns in the '67 registers and write them to magnetic cards. By reading them into the then-brand-new HP-41C, the 41's use of memory could be explored, leading to my talk on same at the 1979 Conference in Santa Clara a month after the '41 came out. As I recall, it took about 20,000+ keystrokes that weekend to map out the data for that presentation.
Yes, I agree with others on the quality of early HP calculator keys!
