In 1976 Mattel released the first microprocessor-based handheld LED game: Auto Race. The next year they released the huge hit Football. Three other games in this lineup were released: Space Alert, Baseball, and Gravity.
Because of its popularity, Football was quickly cloned; the clones were easy to emulate because they used well-documented microcontrollers like Texas Instruments' TMS-1000, National Semiconductor's COP400, and General Instruments' PIC1650. But the Mattel games used unknown Rockwell chips in weird 42-pin quad in-line packages. The only markings on the chips were the Rockwell logo, a date code, and IDs: B6000 for Auto Race, B6001 for Space Alert, B6100 for Football, B6101 for Baseball, and B6102 for Gravity.
Rockwell had produced lots of calculators, many using chips in the QIP 42 package with similar markings, such as A4802, B5000, B5900. They also produced three lines of PPS-4 microcontrollers; PPS-4 required 3 chips for a working system, PPS-4/2 needed 2 chips, and PPS-4/1 had everything in one chip. Some of these were in the QIP 42 package.
Some later Mattel LED games used PPS-4/1 chips, but the pinouts did not match the earlier games. I purchased all the games, many Rockwell calculators, and several PPS-4/1 chips and decapped them to compare the dies. B6000 and B6001 are the same chip with different ROM contents. Likewise, B6100, B6101 and B6102 are the same chip with different ROM. The ROM in the B61xx chips contained 896 bytes, compared to 512 bytes in B60xx. Although the January 1978 issue of Popular Electronics states that the games used PPS-4/1 chips, it was evident from the die shots that this wasn't true. The PPS-4/1 chips and many of the calculator chips were more complex, but some, like B5000 from the 8R calculator were very similar to B6000. B5000's ROM held only 448 bytes, though.
There are patents available for Auto Race and Football and for Gravity. These contain some useful technical information, but the most interesting part of the patents were sentences that said that "a computer program listing" (for Auto Race and Football) and "a computer program object code listing" (for Gravity) were filed with the patents, although these were not included in the patents. I ordered the patent wrappers for these patents from the USPTO. The program listing for Football was actually for a 6502-based prototype. The object code listing for Gravity was just a list of binary numbers. I was able to reorder the bits in the B6102 ROM array to match the object code listing, and I used that same reordering on the ROM arrays in Football and Baseball.
There's also a patent for a Blackjack-playing calculator using a Rockwell chip that includes source and object code. I purchased some of these and decapped the A4821 inside, but it is much more complex and has a larger ROM than the Mattel game chips. I found out that there are two Rockwell calculators that use other A48xx chips: 63R has A4802 and 64RD has A4806. These are the same chip with different ROM contents.
The Electronic Target Game patent mentions the game uses a Rockwell type 6100 microcontroller, but I haven't found anyone who remembers that game.
Scott Stilphen interviewed the creator of four of the games, Mark Lesser. Mark mentions redesigning a Rockwell calculator chip for the games, and only using 511 bytes for Auto Race.
There are many smart people who can analyze dies to figure out exactly how they work, but I am not one of them. Instead I tried to analyze the object code from the B61xx games, comparing them to the PPS-4/1 opcodes (there are 2 varieties of those), and to the A48xx opcodes. I could see patterns, but the sample size was too small for me to figure out the opcodes.
In 2021, Mark Lesser donated a set of notes to the Strong Museum of Play. Along with lots of other information, these contain descriptions of the opcodes and the source code to Auto Race. I wrote a disassembler for the B50xx, B60xx, and B61xx chips, then worked with Hap on the MAME development team. Amazingly, in just a couple of days he had the basics worked out, and was able to figure out the un- and underdocumented quirks of the chips.
reordered ROM dump (padded with empty page 3)
reordered ROM dump (padded with empty pages 12 and 13)
reordered ROM dump (padded with empty pages 12 and 13)
reordered ROM dump (padded with empty pages 12 and 13)