An NES cartridge PCB manufacturing test by Damian Yerrick

This demo detects what mapper it's running on through mirroring, tests how big of a ROM it was written on, verifies that the mapper is mostly working, and displays the result.

Paul Molloy of infiniteneslives.com is selling INL-ROM circuit boards for making NES game cartridges. These boards are configured at manufacturing time to support the mapper circuits required by several games. Among them are Holy Diver by IREM and Batman: Return of the Joker by Sunsoft. Putting those titles together would produce something Burt Ward as Dick "Robin" Grayson might say: "Holy diver, Batman!"

But the Warners probably wouldn't like that. A label in Warner Music Group publishes the Dio album after which Holy Diver was named, and DC Comics (a unit of Warner Bros.) publishes comics starring Batman. In 2010, DC forced NIS America to remove a Batman allusion from the title of the PSP game now known as What Did I Do To Deserve This, My Lord?. So instead, I'm avoiding references to either Warner company's works in the title.

First install Python 3, Pillow, cc65, GNU Make, and GNU Coreutils. The README file for nrom-template describes how to set it up.

The makefile performs four steps:

1. Convert the system font from PNG format to an NES-friendly format
2. Assemble all source code files
3. Link source code files into mapperel-master.nes
4. Generate ROMs with several sizes and mappers

Because several mappers don't guarantee what bank will be visible at power on, ROM generation repeats several parts of the program:

• Every 4 KiB has the Morse code crash handler.
• Every 16 KiB has a trampoline that attempts to change a discrete logic mapper or MMC1 to the last bank.
• Every 256 KiB has the entire test, in order to make testing of SUROM easier.

Once you have test ROMs of the appropriate size, you can use any PRG/CHR splitting program. The contents of all PRG ROMs of a given size are identical no matter the mapper, as are all CHR ROMs. They differ only by header. So go ahead and burn an EPROM for each size and plug it into your socketed boards. However, you can't chop off the end of a larger ROM and use that on a smaller ROM, as it will be missing the special tag that identifies it as the last bank of the ROM. This is intentional in order to prevent builds with disconnected upper address lines from silently working.

• 001: SxROM (Bionic Commando, Final Fantasy, Journey to Silius)
• 002: UNROM, UOROM (7432) (Battle Kid, Contra, DuckTales)
• 004: TxROM (Super Mario Bros. 2-3, Mega Man 3-6)
• 007: AxROM (Battletoads, Jeopardy!)
• 009: PNROM (Punch-Out!!)
• 010: FJROM, FKROM (Famicom Wars, Fire Emblem)
• 011: Color Dreams (Crystal Mines, Spiritual Warfare)
• 028: INL-ROM (A53) (Action 53 volumes 1 and 2)
• 034: BNROM (Deadly Towers, Haunted: Halloween '85)
• 066: NROM, CNROM, GNROM (Balloon Fight, Pipe Dream, Gumshoe)
• 069: JLROM, JSROM (Batman: Return of the Joker, Gimmick!, Hebereke)
• 078.3: IF-12 (Holy Diver)
• 118 TxSROM (NES Play Action Football, Wanderers from Ys)
• 180: UNROM (7408) (Crazy Climber)

The tests look at bank tags, which are increasing numbers spaced at regular intervals in memory that let the program know what bank has been switched into a particular address window. Bank tags are placed at offset 4088 ($FF8) of every 4 KiB bank of PRG ROM and at offset 508 ($1FC) of every 2 KiB bank of CHR ROM or RAM.

First it copies test code into RAM that detects the mapper by how it responds to writes to the supported boards' nametable mirroring ports. Then, because a few mappers react the same way, such as all boards with fixed mirroring, it narrows the field through very basic bank switching tests. This needs to be done in RAM because 32K bank switching may cause the startup ROM bank to be switched out.

After the mapper number is determined, it copies a driver for that mapper into RAM. The driver is responsible for changing 8K CHR banks and WRAM banks and doing detailed tests.

The CHR test starts by asking the driver to switch the first 8K of CHR memory into PPU $0000-$1FFF. If data written to $0000 in CHR space does not change, it's CHR ROM. Otherwise, it's CHR RAM, and the test writes bank tags in CHR RAM banks 31 down to 0. It reads bank tags from the last CHR bank to verify that they're internally consistent. If so, it knows the size of CHR memory.

If CHR ROM is present, the test coompares the font in $0000-$03FF to a copy of the font in PRG ROM to ensure that CHR ROM can be read. Then it verifies all bank tags and saves the result for later.

If it's CHR RAM, it writes a pseudorandom pattern to all of CHR RAM, reads it back, and saves whether it matched for later. Then it repeats the test seven more times with the pattern shifted by a byte each time. While this is going on, it buzzes the speaker to let the user know it hasn't frozen. Finally, it puts the bank tags back and loads the small font into CHR RAM.

Work RAM (WRAM), or PRG RAM, is RAM on the cartridge at CPU $6000-$7FFF. First it looks for the string SAVEDATA at $6100 and sets a flag if it was found. Then the same test is done as for CHR RAM: determine how much, then test each byte. Finally, SAVEDATA is written back. This way, the user can test for battery backup by running the test, powering off, and running the test again. Pressing the Reset button will cause an incorrect result.

Some mappers are more flexible than others in how they map PRG and CHR banks into windows within the CPU and PPU address space. This test steps through all the PRG and CHR bank numbers in each window with various combinations of banking modes. It also checks whether WRAM can be disabled for power-off protection (on mappers that claim to support this) and whether the IRQ works roughly as expected. This is no substitute for an exhaustive mapper-specific test, but it should help determine whether the chips are soldered properly.

Sometimes a test fails hard enough that the framework isn't even sure that it can display the results on screen. When a test fails this hard, your NES will beep Morse code at you. Find a ham and fix your soldering.

• WB (.__ _...):
  Wrong bank at startup. INL's versions of the ASIC mappers guarantee that the LAST 4 KiB of the cart is swapped into $F000-$FFFF at power on. Discrete mappers should be OK as well, as there's a stub every 16K.
• MIR (__ .. ._.):
  The nametable mirroring for this mapper doesn't match any of the supported mappers. Check PA13-PA10, /PA13, CIRAM A10, and CIRAM enable, and don't try running the 78.3 test on an emulator that does not support NES 2.0 format.
• SU (... .._):
  Attempt to switch to second half of SUROM (4 Mbit MMC1) failed.
• LB (._.. _...):
  Mapper detection did not leave the last 16K of PRG ROM swapped in.
• RB (._. _...):
  Attempt to return to the last bank after a test failed.
• CBT (_._. _... _):
  CHR bank tags are not internally consistent.
• FON (.._. ___ _.):
  Font in CHR memory does not match that in PRG ROM. The CHR ROM or CHR RAM is bad.
• DRV (_.. ._. ..._):
  The mapper was detected, but no PRG and CHR test driver for the mapper exists.
• SMS (... __ ...):
  Your mobile phone has received a text message.

The mapper 34 test may freeze in emulators that do not support the NES 2.0 format. They make both the BNROM registers and the NINA-001 registers available at the same time, when the current best practice is to switch between the two based on CHR size.

After the buzzing ends, the following variables are valid:

• cur_mapper: iNES mapper number of the detected mapper
• last_prg_bank: Size of PRG ROM in 4096 byte units, minus one
• is_chrrom: Zero if CHR RAM is present; nonzero if CHR ROM is present
• last_chr_bank: Size of CHR ROM or CHR RAM in 8192 byte units, minus one
• chr_test_result: Zero if bank tags were read correctly from CHR ROM or if all bytes of CHR RAM correctly hold values.
• has_wram: Nonzero if $6000-$7FFF is RAM.
• has_savedata: Nonzero if SAVEDATA was in $6100 of the first WRAM bank at startup
• last_wram_bank: WRAM size in 8192 byte units, minus one; unspecified if has_wram is zero
• wram_test_result: Zero if all bytes of WRAM correctly hold values; unspecified if has_wram is zero
• driver_prg_result: Detailed WRAM and PRG ROM test results
• driver_chr_result Detailed IRQ and CHR test results

All this information is displayed on the screen. The detailed code is 4 digits: WRAM, PRG ROM, IRQ, and CHR ROM/RAM. Zero is normal; anything else reflects something unexpected.

• Mapper 001 1xxx: $E000 bit 4 does not disable WRAM
• Mapper 001 4xxx: $A000 bit 4 does not disable WRAM (SNROM), or $A000 bit 4 disables WRAM (all but SNROM)
• Mapper 004 2xxx: Read-only mode not present

In iNES format environments that don't support NES 2.0, such as FCEUX and PowerPak, the MMC3 test will return a warning about lack of write protection on WRAM. These environments don't implement this feature because not implementing it gets the two StarTropics games, which use MMC6, to run with the MMC3 driver.

Some INL-ROM products implement only a subset of the entire mapper as a cost-saving measure. For example, MMC3 may lack WRAM protection or may fix C and P values. Detailed codes will reflect unimplemented features. If a subset changes mirroring behavior, the mapper may not be recognized in the first place.

For the convenience of users testing large numbers of boards without looking at the screen, the result is also beeped through the speaker with one note per nibble so that an attentive ear can pick up any deviation from the intended melody.

0: Bb       4: F        8: Mid C    C: G
1: Low C    5: G        9: D        D: A
2: D        6: A        A: E        E: B
3: Eb       7: Bb       B: F        F: High C

There are three groups of nibbles:

1. Tweet, mapper number (2)
2. Tweet, PRG ROM size in 32768 byte units (2), PRG RAM size in 8192 byte units (2), buzz if PRG RAM defective, ding if battery, PRG RAM detailed result (1), PRG ROM detailed result (1)
3. Tweet, CHR size in 8192 byte units minus 1 (2), CHR RAM or ROM (1), buzz if CHR RAM or ROM defective, IRQ detailed result (1), CHR detailed result (1)

Not all mappers are supported.

NROM and CNROM are detected as mapper 66, but the board name should be correct if their memory size matches a known configuration.

The IRQ test measures only gross behavior of the interval timer of the MMC3 or FME-7. It is intended to measure continuity of the IRQ pin during board assembly, not as a substitute for detailed tests of exact timing of a CPLD reimplementation of a particular mapper.

The "Flash ID" results mean nothing. They are reserved for future expansion. Pull requests are appreciated.

Copyright 2013-2017 Damian Yerrick

Available under zlib License.

Not sponsored or endorsed by DC Comics, IREM, Nintendo, or Sunsoft.