A game cartridge holds ROM chips wired straight to the console’s address and data lines. The processor reads its program from them exactly as it would from memory inside the machine, so nothing has to load: switch on, and the game is already there.
The less obvious consequence mattered more. Because the cartridge sits on the processor’s bus, it can hold more than memory. A publisher could put almost anything the console’s designers had left out into the slot: logic to switch banks of ROM, RAM with a battery to keep saved games, extra sound channels, even a second processor faster than the console’s own. A cartridge console could be upgraded one game at a time.
Fast facts
- First: the Fairchild Channel F, shown in June 1976, which ANALOG Computing called “the first programmable video game system”.
- What is inside: program ROM; on many consoles graphics ROM or RAM; and sometimes a mapper, save RAM with a battery, or a co-processor.
- Growing past the address space: bank switching, from the Atari 2600’s 8K games to the NES’s MMC1 and MMC3.
- Co-processors: the Super FX and SA-1 on the Super NES, and Sega’s SVP for Virtua Racing on the Mega Drive.
- Drawbacks: mask ROM is expensive and cannot be changed after manufacture.
The Channel F’s Videocarts
Fairchild’s console, shown at the Consumer Electronics Show in Chicago on 15 June 1976, took its games on cartridges it called Videocarts. They were deliberately shaped like eight-track music tapes, and the manual told owners to insert them in “much the same way an eight-track audio cartridge is into a cartridge tape player”. ANALOG Computing summed up the change in 1988: Fairchild “introduced the first programmable video game system, the Channel F, in 1976”, and “transferred the game programs from the console to interchangeable game cartridges”. Until then a home video game console played the games built into it and nothing else.
The idea came out of the processor. Jerry Lawson’s team built the console round Fairchild’s own F8, which came as a pair of chips: a 3850 processor and a 3851 program storage unit holding 1K of ROM. Retro Gamer’s Videogames Hardware Handbook describes a Videocart as containing “almost no circuitry – just two of the companion ROM chips”. Richard Olney, who worked on the F8’s design, wrote that its two-chip design, with the program fixed into the 3851 when the chip was made, “made the F8 a good choice for a cartridge based game system”. The two 3851s inside the console held the built-in games and “also provide subroutines that are used by the cartridge based programs”. According to the Handbook, removable game memory “hadn’t been done before” because of the risk of sparks and static, so the US Federal Communications Commission made Fairchild submit every new cartridge for testing, and the console, the cartridge bay and each cartridge had to be shielded in metal.
More game than the address space
Every cartridge machine hit the same limit sooner or later: games outgrew the part of the address space the processor could give them. The Atari 2600’s 6507 processor can see only 4K of cartridge ROM at a time. Electronic Games explained to a reader in November 1982 that from Asteroids and Missile Command on, the console “is capable of up to 8K, using a special bank switch that flip-flops two 4K programs”. The cartridge holds more ROM than the processor can see, and logic on the cartridge decides which part is visible.
The NES made that logic a product line. The console gives a cartridge the processor’s upper 32K, $8000-$FFFF, for program ROM, and its picture chip, the PPU, reads its tile patterns straight from the cartridge too, so the graphics never pass through the processor. A game that fits in 32K of program and 8K of graphics, as Super Mario Bros. does, needs nothing else. Bigger games carry a mapper chip. Nintendo’s MMC1 of 1987 switched up to 256 KB of program in 16 KB banks and let a game choose its own screen mirroring; the MMC3 of 1988 switched 8 KB banks, reached 512 KB, and added an interrupt part-way down the screen, which Nintendo Power credited for Super Mario Bros. 3’s fixed status bar. Konami designed its own mappers, among them the VRC2, which switched graphics in eight 1 KB pieces.
The Japanese Famicom’s cartridge slot also has an audio input, which the Western NES dropped, so Famicom cartridges could add sound hardware. Konami’s VRC6 gave Akumajō Densetsu two extra pulse channels and a sawtooth wave; its Western release, Castlevania III, had to make do with the console’s own sound.
A second processor in the box
The Super NES took the idea furthest. Nintendo’s development manuals document three chips that could ride in the cartridge: the DSP-1 maths processor, the SA-1 and the Super FX. The Super FX, designed with Argonaut, was a graphics processor that ran, in Nintendo’s words, “six times as fast as the Super NES CPU”, and it drew the polygons of Star Fox (Starwing in Europe) in 1993. Argonaut told Edge in 1994 that the faster Super FX 2 was rated at 21.477 MHz. The SA-1 was a 65C816 like the console’s own processor, running at 10.74 MHz, “four times faster than the normal operating speed of the Super NES CPU”, alongside it; Edge reported that Super Mario RPG used one to unpack the compressed graphics that filled its 32-megabit cartridge.
Sega answered with the SVP chip in the Mega Drive version of Virtua Racing. Yu Suzuki, whose team made the arcade original, told Edge in 1994 that the conversion was attempted “mainly because the SVP chip meant it wasn’t an over-complicated conversion. Without it, I don’t think I would have even attempted it.” The approach had a cost. Argonaut pointed out that the Super FX ran its program from the game’s own ROM, so one chip design served every game, while the SVP appeared to carry its own program in masked ROM, so that “every new game needs a new SVP chip”.
Saving the game
A cartridge could also remember. RAM on the cartridge, kept alive by a lithium cell, held saved games. Nintendo’s MMC1 specification of 1987 named a CR2032 for the job, and the cartridge release of The Legend of Zelda kept three saved games that way. Nintendo Power explained the point to its readers in January 1991: before battery back-up, a player who wanted to finish a game “had to do it during one play session”. It reckoned a battery would keep the data “about five years”. Owners of games on Nintendo’s newest mapper, the MMC5, were promised that they “won’t have to push RESET on your Control Deck while pushing POWER when you want to quit”.
What cartridges cost
A cartridge’s program is fixed when its mask ROM is made, in a factory and in quantity, and a bigger ROM costs more on every copy sold. Edge described Nintendo’s Famicom Disk System of 1986 as “an attempt to counter rising ROM costs”, with disk games at roughly half the price of cartridges. Within a few years cheaper ROM and battery saves had brought publishers back to cartridges.
The price never went away. When the Game Boy Advance arrived in 2001, Edge found some publishers refusing to develop for it “owing to the low margins offered by Nintendo’s cartridge pricing scheme”. Martin Merren of Crawfish Interactive told the magazine what that meant for the game itself: “we might get squeezed on the amount of cartridge space that we can use, and the possibility of battery back-up.” The Nintendo DS of 2004 kept a Game Boy Advance slot, but Edge reported that its own software would come on neither cartridges nor discs, “but a new ‘thin memory card’ format”.
The slot as a gate
Because a cartridge is hardware, the company that controls its manufacture controls the platform. The NES carried a lockout chip, the 10NES: a matching chip in every licensed cartridge had to answer it, or the console would not run the game. Compute!’s Gazette put the consequence plainly in 1990: “Since Nintendo is the only producer of this chip, it’s able to control the quality of the games produced for the NES.” The Famicom had no such chip. Nintendo also built the cartridges, and its licensing rules are described in the Nintendo Seal of Quality entry.
The paperwork reached inside the ROM. Nintendo’s Game Cassette Registration Specification of 18 February 1987, an unpublished document for licensees, required every cartridge to carry a registration area near the top of memory, from $FFF0, holding a maker code, a registration number, the cassette type and memory size, checksums over the program and the graphics data, and the game’s registered title. The Super NES’s SA-1 carried the control into the co-processor: Nintendo’s manual notes that SA-1 cartridges use a different ROM from standard game paks, which “guards against unlicensed products”.
Home computers
Home computers used the same idea for different reasons. On the Commodore 64, a cartridge at $8000 starts itself if it carries the five bytes $C3 $C2 $CD $38 $30 (the letters CBM, inverted, and the digits 80) at $8004: at power-on the machine checks for them and jumps to the address the cartridge gives. The same expansion port took hardware that was not a game at all, such as the Action Replay freezer cartridge, which worked because the port is wired to the processor’s bus.
Sources
- Retro Gamer: Videogames Hardware Handbook volume 1 (2nd edition, 2016), “Fairchild Channel F”, pp. 18–21: Jerry Lawson and the F8; the 3850 and 3851; Videocarts and the eight-track design; the contents of a cartridge; FCC testing and shielding; the show debut on 15 June 1976.
- ANALOG Computing issue 60, May 1988, p. 84, “Video Game Digest”: the Channel F as “the first programmable video game system” and its interchangeable cartridges.
- Richard Olney, “F8 Microprocessor Design” (2020), a designer’s account: the two-chip design and its fit for cartridges; the console’s 3851 chips holding the built-in games and routines used by cartridge programs.
- Electronic Games vol. 1 no. 9, November 1982: 8K Atari 2600 games and bank switching.
- NESdev Wiki, “MMC1” and “MMC3” (community reference, web, secondary), as cited in the MMC1 and MMC3 entries: the NES cartridge map and mapper capacities. https://www.nesdev.org/wiki/Mapper
- NESdev Wiki, “VRC6 audio” (community reference, web, secondary): the VRC6’s channels and Akumajō Densetsu; the Famicom’s cartridge audio input and its absence on the NES. https://www.nesdev.org/wiki/VRC6_audio
- Nintendo Power volume 20, January 1991, pp. 29–30: the MMC3’s timer and Super Mario Bros. 3; “one play session”; battery life of “about five years”; the MMC5 and the RESET button.
- Nintendo, Development Manual, Book II: the SA-1’s 65C816 core, its 10.74 MHz clock and the protection against unlicensed products (section 1); the Super FX, as cited in the Super Nintendo entry.
- Super Play Gold, 1993: the Super FX in Starfox/Starwing.
- Edge issue 7, April 1994: Argonaut on the Super FX 2 at 21.477 MHz and on the SVP’s masked ROM.
- Edge issue 9, June 1994: Yu Suzuki on the SVP chip and Virtua Racing.
- Edge issue 32, May 1996, p. 39: Super Mario RPG, the SA-1 and its 32-megabit cartridge.
- Edge issue 54, January 1998: the Famicom Disk System, “rising ROM costs” and disk game prices. Retro Gamer: Videogames Hardware Handbook volume 2, 2010, pp. 220–225, as cited in the Famicom Disk System entry: the return to cartridges as ROM and battery saves became cheaper.
- Edge issue 102, October 2001, pp. 48–52: Game Boy Advance cartridge pricing; Martin Merren on cartridge space and battery back-up.
- Edge issue 134, March 2004: DS software on a “thin memory card”.
- Compute!’s Gazette, April 1990: Nintendo’s proprietary cartridge chip.
- Nintendo, MMC Cassette Programming Specification (18 November 1986, revised 26 March 1987) and Game Cassette Registration Specification (18 February 1987), unpublished licensee documents: the CR2032 cell; the registration area and its fields. Every other statement in this entry stands on the public sources.
- Sheldon Leemon, Mapping the Commodore 64 (COMPUTE! Books, 1984), entry for location 32768 (
$8000): the autostart cartridge signature.