Professional software development, amateur BMW tinkering, old arcade game stuff


Konami Track & Field arcade pcb repair

Booted up to a screen full of 0′s.  That meant the CPU was partially running as it was clearing the text layer to consistent 0 instead of leaving it as random garbage.  The program ROMs were tested on PC and 2 of the 5 gave inconsistent reads.  2 replacement eproms were burned and game started up!


There was a further graphics problem – corrupt text on the hi-score names in game.  At first I thought this might be a problem with the graphics ROMs and indeed 1 gave inconsistent reads, but replacing this didn’t fix the problem.  Some Googling revealed this is a common problem when the battery on the board dies – it seems the game then reads garbage data from the save memory and doesn’t do any validity checking on it – so it just copies garbage chars to the screen.  A fresh battery was installed and the save memory reset via the dip switches.





Taito Legend Of Kage arcade pcb repair

Much like the earlier Tiger Road repair every other line missing in the sprites suggests a double buffered sprite line buffer system, and that one of the buffers have failed.  Handily the sprite ROMs were in sockets already on this board, so I removed them one at a time and booted the board – removing all except IC21 caused additional visual faults, so good chance this was the culprit.


Replacement RAM was fitted and board was 100%.

IMG_5760 IMG_5761




TAD Cabal arcade pcb repair

Like the Legend of Kage and Tiger Road repairs all sprites had every other scanline missing suggesting a failure in double buffered sprite line RAM.


The entire top board is for sprite processing and contains 12 ram chips (the tilemap, palette and main RAM is all on the bottom board).  The bad chip was found by piggy backing a known good RAM chip on top of each RAM until the problem went away.

IMG_6157 IMG_6158

Trackball to joystick mod was also built and installed thanks to the work at




Konami X-Men arcade pcb repair

This pcb was a mess of previous repairs and missing parts – two crystal oscillators missing, program ROMs missing, audio mask ROM missing, various TTL and RAM chips replaced.  So clearly many problems in the past and someone likely tried to repair it, failed and then harvested parts to fix other boards.

I pulled some crystals from a parts board (Konami Run and Gun) and burned some fresh program ROMs from the MAME set.  Game booted to a solid yellow screen.  Like the Vendetta repair I suspected palette RAM and when new RAM was piggybacked on top the usual Konami check screen came up with a bunch of errors.


Eeprom @ 13B error

13B is the eeprom used for storing game configuration data – when I tested this with a logic probe there was no data output – but actually there was no +5V to it either…  it seems this was a victim of a previous repair attempt – it had been removed and replaced, but had a bad solder joint on the +5V line.  With that corrected the game still reported an error – and this time the data pins were active.  Schematics are available for this part of the board and there was no continuity between the two outputs of the eeprom and where they should go – trace damage from the previous repair attempt.  Patch wires soldered on the bottom of the board – and success!  The screen was no now longer upside either as the game stored the flip screen configuration data correctly.

The remaining errors all related to the audio section of the board.  Like Vendetta I decided to patch the program to skip errors and see what happened.



Broken Colour Output Custom

Well, the game ran surprisingly well!  All graphics and inputs were working, just no sound, and clearly no blue colour in any graphics.  There are schematics for the video output, which consists of two LS273 latches and a bunch of LS07′s.  Logic probe seemed to show this was all working correctly, so the problem was likely the custom 052535 IC – a replacement was taken from a Run n Gun and the blue channel restored.



Audio errors

Errors 6F, 7G, 4E, 2D are all in the audio section of the board – program ROM, two RAM chips and the missing mask ROM.  Piggybacking RAM had no effect, and the program ROM was freshly burned from MAME so it should have been correct.  Further probing revealed the Z80 CPU was not even running – no activity on data or address lines – so it makes sense the main CPU reported all the audio parts bad.

Before declaring the Z80 dead I checked the control lines – and the WAIT line was being held – so something was actually forcing the Z80 not to run. No schematics are available for this section so I traced the WAIT line to the output of a LS08 chip  – that’s just an AND gate – and eventually traced the input to the gate from the custom audio chip.  So some info not in MAME is that the audio chip actively controls the Z80 wait, probably when accessing shared memory.

I temporarily removed the LS08 so the Z80 could run and there was activity on the data and address lines so the CPU itself was probably good.  At this point I stalled for quite a while with no ideas.  After finding the bad Konami PAL on the Vendetta repair though I checked the X-Men audio PAL and it was completely failed.  When replaced some synth sounds could be heard during the game!  The WAIT problem turned out to be the custom chip was not getting a clock signal – the Run N Gun dual frequency crystal was bad – so one was taken from a working game (Premiere Soccer) and the LS08 replaced – this finally cleared up all the audio errors in the self check.

Sample ROM

Final problem was the missing sample ROM – a 2 megabyte 27c160 eprom can be used here as replacement.  Lots of traces on the board had been damaged but all address/data lines are shared with the custom audio chip so a logic probe in the socket will show if any lines are disconnected.


Unusual to see a board with so many unrelated problems, but fixed 100% at last.


IMG_5503  IMG_5515





Konami Vendetta arcade pcb repair

Game booted to a solid white screen.  Often this can be failed palette RAM as if it fails and all outputs stick high this is usually interpreted as white for every pixel.  Spare RAM was piggy backed onto the palette chips at 7F and 8F and success – some garbage displayed!


Amongst the garbage was a string about bad RAM at 16C- this is the main program RAM.  Another chip was piggybacked to replace the failed one and the familiar Konami check screen came up.


At this point though the repair slowed down a lot – bad RAM was reported at 16G and 16F – this is the tilemap ram.  4F and 5F is the audio RAM and ROM.  Now the audio ROM checked out ok compared to the MAME set, and piggybacking RAM made no difference.  I wasn’t convinced 16F/16G were actually bad either – as if they were the check screen itself should be corrupt.

Eventually I modified the program ROM to continue past the check screen (simple NOP for the branch that resets after the test – quickly found with the MAME debugger).  And despite the errors – the game ran pretty good!  Sound was 100% correct and the only graphics glitch I could find was the high score table.  A weird glitched also existed where pushing up on the joystick started the service menu and pushing down made all sprites disappear.  Inside the service menu the mask ROM test reported every ROM as bad – even though they clearly were not.




The MAME source code had some interesting info:


PORT_BIT( 0×01, IP_ACTIVE_HIGH, IPT_UNKNOWN ) PORT_READ_LINE_DEVICE_MEMBER(“eeprom”, eeprom_serial_er5911_device, do_read)
PORT_BIT( 0×02, IP_ACTIVE_HIGH, IPT_SPECIAL ) PORT_READ_LINE_DEVICE_MEMBER(“eeprom”, eeprom_serial_er5911_device, ready_read)
PORT_BIT( 0×08, IP_ACTIVE_HIGH, IPT_CUSTOM ) PORT_VBLANK(“screen”) /* not really vblank, object related. Its timed, otherwise sprites flicker */


So joystick up on the 1 player input port is bit 0×4 – same as the service bit 0×4 on another port.  Joystick down is bit 0×8 and that same bit affects sprites.  So it’s clear those are being mixed together somehow.  The culprit turned out to be an LS04 chip that the schematics show controls the enable pins on some input buffers.  Probably the player 1 input buffer and the service/sprite buffer both enabled at once.



So the final problem on the pcb took ages to find – literally the last thing I checked were the two custom programmed PAL’s that are used to form the CPU memory map.  It seemed very unlikely there was any problem there given how much of the game worked fine.  Whenever a replacement GAL was programmed and piggybacked on top of P2 the RAM/ROM check errors went away and the high score table was fixed!  An extremely unusual fault.  The PAL clearly worked 99% of the time until some timing edge case occurred.






Konami TMNT arcade pcb repair

Board #1


Game played, but with an overlay of text over everything.  I suspected the tilemap custom chip wasn’t reading from tilemap RAM properly but nothing seemed obviously wrong until I found some bad traces on the bottom of the board.  Someone had previously tried to repair them, and it had probably worked at the time but now the thin lines had corroded further.  A patch wire on the top of the board was the safest fix.

IMG_5294 IMG_5304


Board #2


Board booted to consistent screen of garbage text.  I had a hunch this was actually the ROM/RAM check screen – but with the graphics addressing broken in some way so incorrect text was displayed.  Logic probe check on tilemap related chips (found via schematics) found a dead input to the LS139 that controlled VRAM chip select.


Pin 3 (A16) comes direct from the main CPU and with this floating it caused garbage to be loaded to VRAM.  I couldn’t find any broken traces, but there must have been one somewhere.  A wire patch was added underneath the board direct to A16 on the 68000 and everything worked.






Technos Double Dragon arcade pcb repair

Board played but no sound at all.

First thing was to check the amplifier – there was a faint humming sound from the speaker and the hum changed as the volume control was moved, so the power amp part at least seemed to be working.

Next thing is to check the digital side – a logic probe showed the sound CPU pulsing as well as the ROM and various address and data pins, so it was definitely executing code.  A good trick here and on similar games is to check if the YM2151 IRQ pin is pulsing – this is because the YM2151 will only send out interrupt requests if the Z80 correctly programs it.  If the CPU is just executing garbage then it’s unlikely the YM2151 will get the correct setup commands.

In this case – the logic probe showed the IRQ pin was stuck high.  More probing showed the RAM chip enable (CE) pin never pulsed either.  So it seemed the CPU program had probably crashed because it could not access the RAM.  Schematics showed CE is driven from the LS138 at IC79 and indeed amongst the dirt there seemed to be some water damage/corrosion!

The LS138 turned out to be fine though a logic probe showed pin 1 wasn’t connected to anything.  Schematics show it should connect to the Z80 address bus – in fact this trace had corroded so I soldered a patch wire on the board underside.  The YM2151 IRQ started pulsing and music worked!


Footstep samples however just played a loud static noise – this game has two MSM5205 sample players.  These take 4 bit data, so a pair of LS157 multiplexers convert the 8 bit ROM data into two 4 bit halves to feed the sample player chip.  One of the LS157 output lines also went through the corroded area which meant a dead input into the MSM5205.  The line was patched and all sounds worked again.





Taito Chase HQ arcade pcb repair #2

Two different boards with pretty much the same fault – some kind of horizontal glitch in all the sprites with blocks of 4 pixels appearing out of place.

IMG_5359 IMG_5365

At first I thought this would just be a simple addressing problem affecting the sprite ROMs but when I examined the data format in MAME each 16 pixel row of each tile is stored at a unique address (so the 16 pixels map directly to the 16 bits on the ROM data bus).  So if the glitch is within a 16 pixel block it can’t be addressing, it has to be elsewhere.  Unfortunately the problem wasn’t easy to find as the entire video board is for sprite processing.  Eventually I found it – dead outputs on the PROMs at IC92 or IC94 can cause this.


Only one PROM of the pair was bad and it was replaced with one from a parts board.  Problem solved.

IMG_5376 IMG_5392


Capcom Tiger Road arcade pcb repair

This game had every second scanline missing in all the sprites.

IMG_5215 IMG_5217

A lot of arcade game games work by using two line RAM buffers for sprites – the sprite logic builds line N+1 in RAM0 whilst line N in RAM1 is displayed to the screen, then they swap and RAM0 outputs to the screen whilst the new line is built in RAM1 and so on.

So a missing scanline in sprites suggests one of the RAM chips in the double buffer has failed.  By piggybacking good RAM onto the chips on the video board I found the fault right away.  The lower line buffer RAM had failed.  A new one was soldered in and game fixed.

IMG_5222 IMG_5227




Rare Battletoads arcade pcb repair

(One of the most complex/time consuming repairs I’ve done – so a very detailed write-up!)

Part 1

The board initially showed no signs of life with most TTL on the board not active at all.  Digging deeper showed there would be tiny pulses of activity then nothing for a few seconds, then repeat, so it seemed a watchdog was constantly trying to reset the CPU.  This was confirmed on the /RESET line of the CPU which was being pulled low for a period then released high as expected.  Voltage to the CPU was good, as were the clock signals (the TMS34020 takes two clocks, one regular CPU, one ‘video’ clock).

The first thing the CPU is meant to do on reset is to fetch the reset vector from program ROM so it can start executing code.  So examining the program ROMs showed the address lines were all pulsing at reset as if the CPU was trying to read, but the ROMs never output any data.  The reason why no data was output is because the /OE (output enable) line is never set.  There are no schematics available for this game, so to figure out why requires a lot of tracing and reverse engineering.  Time to go deep..

Program ROM /OE is driven from pin 19 of the programmable GAL chip at U144.  I dumped the GAL to equations, and also did the same with the MAME dump – both match so I can rule out the GAL itself being bad.  The equations are:

/o21 = i7 * i8 * /i9
/o20 = i23 * i7 * /f17 * i8 * i9 * /i10 * i14 * /i11
/o19 = i23 * i7 * /f17 * i8 * i9 * /i10 * /i11
f17 = i3 * /i4
o16 = vcc
/o15 = i7 * /f17 * i8 * i9

So the outpin on pin 19 (/o19) is a function of 9 different inputs.  Ouch.  Without knowing what they do yet, I can see that all except i10 pulse at reset.  Time to trace where they all go, which is very time consuming with the tiny traces on this board.

i7-i9 are effectively bits 29-31 of the CPU address bus (they get latched into the LS373@U139 first).  It makes sense these bits are involved, because if we reference the memory map from MAME – those top 3 bits are used to identify each ‘section’.

static ADDRESS_MAP_START( main_map, AS_PROGRAM, 16, btoads_state )
AM_RANGE(0×00000000, 0x003fffff) AM_RAM
AM_RANGE(0×20000380, 0x200003ff) AM_READWRITE(main_sound_r, main_sound_w)
AM_RANGE(0×20000400, 0x2000047f) AM_WRITE(misc_control_w)
AM_RANGE(0×40000000, 0x4000000f) AM_WRITENOP    /* watchdog? */
AM_RANGE(0×60000000, 0x6003ffff) AM_RAM AM_SHARE(“nvram”)
AM_RANGE(0xa0000000, 0xa03fffff) AM_READWRITE(vram_fg_display_r, vram_fg_display_w) AM_SHARE(“vram_fg0″)
AM_RANGE(0xa4000000, 0xa43fffff) AM_READWRITE(vram_fg_draw_r, vram_fg_draw_w) AM_SHARE(“vram_fg1″)
AM_RANGE(0xa8000000, 0xa87fffff) AM_RAM AM_SHARE(“vram_fg_data”)
AM_RANGE(0xa8800000, 0xa8ffffff) AM_WRITENOP
AM_RANGE(0xb0000000, 0xb03fffff) AM_READWRITE(vram_bg0_r, vram_bg0_w) AM_SHARE(“vram_bg0″)
AM_RANGE(0xb4000000, 0xb43fffff) AM_READWRITE(vram_bg1_r, vram_bg1_w) AM_SHARE(“vram_bg1″)
AM_RANGE(0xc0000000, 0xc00003ff) AM_DEVREADWRITE(“maincpu”, tms34020_device, io_register_r, io_register_w)
AM_RANGE(0xfc000000, 0xffffffff) AM_ROM AM_REGION(“user1″, 0)

As an example you can see output 21 on the GAL has this equation:

/o21 = i7 * i8 * /i9

So if i7 is set (0×20000000) and i8 is set (0×40000000) but i9 is NOT set (0×80000000) then the resulting address is 0×60000000 which means ‘nvram’ is being accessed in the map.

Back to output pin 19..  i3 and i4 are effectively address bus bits 2 & 3 (latched via ls373@u7).  i11 goes straight to pin 8 on the CPU which is /RAS.  i10 goes to CPU pin 133 which is ‘/TR /QE’.  So if we refer back to the equation:

/o19 = i23 * i7 * /f17 * i8 * i9 * /i10 * /i11

For /o19 to be low (and enable the program ROM) we need the address lines to be correct, /i11 (RAS) to be low which it always is, but also ‘/TR /QE’ to be low, which always measures as high.  The manual describes ‘/TR /QE’ as ‘Transfer/output-enable.  During a local-memory read cycle, TR/QE
functions as an active-low output-enable to gate from memory to LAD0–LAD31.‘.  So it makes sense it’s involved, it’s part of the cycle of reading, but why is it stuck?


At first I thought the CPU had just failed internally – but there are some control pins that can affect the CPU state.  The bus fault (BUSFLT) pin checked out ok, and the CPU wasn’t stuck in host mode (HCS pin).  The smoking gun was a control pin called LDRY – this can halt the CPU in the middle of a memory cycle.  This traced back to pin 21 of the GAL @ U11.  Again, I worked out the equations for this GAL and started to trace what controls pin 21.  The actual problem though was that this GAL had just failed and the outputs were garbage.  Replacing this GAL fixed the LDRY input and the CPU finally started to read from the ROMs!

Part 2

So despite the CPU seeming to be running the game was still stuck on a black screen and resetting every second or so.  The CPU was definitely running some code at least as video sync was now working (I assume the CPU sets up the video output).  I then spent a lot of time tracing all the data and address lines, and found no problems.  Eventually I decided to try and disable the watchdog after noticing the game in MAME takes a couple of seconds on a black screen before booting up and the pcb seemed to reset after only one second.

The watchdog/CPU reset is controlled by an IC called LTC690 at U83.  A logic probe didn’t show anything wrong with this – there were occasional writes to reset the watchdog, but probably not as many as there should have been.  The datasheet states the watchdog can be disabled (but the CPU reset functionality kept) by floating the watchdog input.  So without much to lose I just cut the trace (pin 6).  And the game worked!


I have no idea why the watchdog circuit has failed – but it’s arguably not needed as the watchdog is really just to protect against software faults and keep a machine running, so there’s nothing wrong with leaving this disabled for a ‘home’ pcb.


Part 3

On the home straight now – graphics and inputs fully working, but no sound.  Sound comes from the BSMT2000 custom chip – a logic probe showed it had a clock input but no activity on the data or address bus.  I was hoping the fault was that it was not receiving any data from the Z80 rather than having failed itself as I had no replacement.  The Z80 memory map is controlled by a GAL like the main CPU.  As before I dumped the equations from the MAME version which showed what outputs were in use.  With this info and the logic probe it was pretty clear to see the GAL inputs were working, but the outputs were all stuck.  I piggy-backed a replacement GAL and sound worked!  So indeed the fault was that the Z80 was unable to write data to the BSMT2000.


The bad GAL was desoldered and the replacement installed and the board was back to 100%.