Whence Came the Famicom’s Brain?

The MOS Technology 6502 was a landmark microprocessor. It was powerful, easy to program, and cheap, debuting for an astounding $25 in 1975. In its wake, competitors scrambled. Microprocessor prices dropped dramatically, but the damage was already done. MOS’s bargain basement chip was overwhelmingly attractive to PC and videogame manufacturers: Atari, Commodore, and Apple, among others, launched successful videogame and PC platforms built around the 6502.

But why did Nintendo choose the 6502?

David Sheff’s Game Over (1993) is widely regarded as one of the best sources on Nintendo’s corporate history. Rightly so. He had unprecedented access to the company at a time when they were much less reticent to talk to the (Western) press. And he weaves a compelling, personal tale, exposing the human side of a now-monumental tech company.

By his account, Nintendo chose the 6502 based on its even balance of power and price. Masayuki Uemura, Nintendo’s head of the hardware-focused Research and Development Team, R+D 2, grappled with the problem for many months, consulting with the company’s arcade engineers to try and figure out how he might capture their magic in a home console:

Uemura spent eighteen-hour days with the arcade engineers trying to determine the essence of the key components to the circuitry in the best coin-operated games. Only that essence could be carried over to the central processing unit of the new system. Finally he chose a relatively standard microprocessor called a 6502… (30)

That first sentence is quite a mouthful. ‘The essence of the key components to the circuitry’ even has an air of mysticism to it. What wondrous alchemy hath these arcade engineers wrought?! The forced syntax and tone of wonderment points to a flaw in Sheff’s research that I think is often overlooked. I get the sense that he’s really struggling with the technical side of his subject. It’s understandable. Sheff was not a computer journalist–he sort of fell into the field due to his young son’s interest in video games. But it makes me suspicious of the details.

Just one paragraph later, he writes, ‘Other companies’ game machines used an integrated circuit made for old-model personal computers, Texas Instruments’ T19918, which allowed six to eight colors.’ I think he really means the Texas Instruments TMS9918, a video display controller that found its way into the ColecoVision, Sega SG-1000, and the MSX. Sheff’s color count is also incorrect, though there is some admitted variation due to the screen mode and chip revision used. In any case, the available palette was sixteen colors, which both the SG-1000 and ColecoVision supported.

Granted, I can play expert in 2012 and quickly browse Wikipedia for a bevy of early console tech specs. Sheff didn’t have the same luxury in the early 1990s and may not have thought these details to be important to his overall tale. He’s probably right, but the point still stands. What else might he have missed?

Reading Brian Bagnall’s On the Edge: The Spectacular Rise and Fall of Commodore (2006), I came across this interesting passage, which I’ll quote at length:

[Commodore 64 programmer] Robert Russell investigated the NES, along with one of the original 6502 engineers, Will Mathis. “I remember we had the chip designer of the 6502,” recalls Russell. “He scraped the [NES] chip down to the die and took pictures.”

The excavation amazed Russell. “The Nintendo core processor was a 6502 designed with the patented technology scraped off,” says Russell. “We actually skimmed off the top of the chip inside of it to see what it was, and it was exactly a 6502. We looked at where we had the patents and they had gone in and deleted the circuitry where our patents were.”

Although there were changes, the NES microprocessor ran 99% of the 6502 instruction set. “Some things didn’t work quite right or took extra cycles,” says Russell. […]

The tenacity of the Japanese was obviously formidable. Russell offers an opinion on why the Japanese elected not to purchase chips from North American sources. “They looked at the patents and realized that we weren’t going to let them come over and sell against us,” he says. (467-8).

To put this information in context, let’s think about Nintendo’s attempted forays into the U.S. home console market. Initially, Nintendo didn’t plan the Famicom’s Western debut as a solo affair. In 1983, Nintendo and Atari were in negotiations for the latter to license and manufacture the Famicom in the United States. At the time, Atari was hoping for a successor to the underwhelming 5200. They were simultaneously courting two suitors in an effort to pick the best candidate: General Computer’s MARIA chip and Nintendo’s near-final Famicom hardware. A letter detailing Atari’s visit to see the Famicom prototype shows that Atari was stringing Nintendo along until they had a better idea of the MARIA’s cost and efficiency. If they chose to partner with Nintendo, there were a number of stipulations, including Atari’s lack of access to the Famicom’s CPU/PPU specifications and Nintendo’s explicit control over all software development.

So why the secrecy surrounding the Famicom’s internals? Certainly part of Nintendo’s reluctance was the lack of a finalized deal. But what if there were details they didn’t want to disclose to a potential partner, like a chip modification that might get Atari into legal trouble with Commodore?

This is all speculation, but it makes you wonder. If Uemura settled on the 6502 as a cost-efficient but powerful microprocessor for Nintendo’s console, why would they try to hide or disguise it by scraping away the patentable bits?

Keep in mind that Commodore was a dominant force in the PC market at the time, but they were having little luck in Japan. There, Commodore withered against the force of dozens of Japanese PC manufacturers that opted to supported the MSX standard. Commodore’s CEO Jack Tramiel knew Japan was a lost cause, but he intended to keep Japanese corporations out of the U.S. PC market at all costs. Bagnall has a telling quote:

In a December 1982 broadcast of the Computer Chronicles, Jack Tramiel told his hosts, “As far as the Japanese are concerned, I was able to keep those people out of the US market and almost the world market for the past seven years.” […]

Host Gary Kildall agreed, saying, “The C64 was definitely one of those devices that kept the Japanese machines out.” (297)

So Commodore is headed by a fiercely competitive, anti-foreign entrepreneur who aims to keep the Japanese at bay. That alone could pose problems for Nintendo. But Commodore had also purchased MOS Technology in 1976 and consequently owned the license to the 6502. Can you imagine Nintendo trying to negotiate a manufacturing deal with Commodore in 1982, the same year that Tramiel expressed his distate of the Japanese?

Faced with that opposition, a bit of engineering subterfuge may have been understandable. Japanese manufacturer Ricoh handled Nintendo’s custom 6502 modifications, so Commodore couldn’t really oversee its production. By the time the NES could be reverse-engineered by Commodore’s programmers, it was already too late. Nintendo had arrived.

The question is why Nintendo didn’t opt for the other popular, capable, and low-cost microprocessor, the Zilog Z80? It ran a number of Nintendo’s competitors’ platforms, like the Sega Master System and Game Gear, but also Nintendo’s very own arcade hit Donkey Kong. Likewise, Nintendo’s first U.S. licensee, Coleco, used a modified Z80 for the ColecoVision and their port of Kong was lauded as the most accurate home console translation of its time.

I don’t have the answer, but there’s some evidence that Nintendo very nearly did go with the Z80. They were undoubtedly impressed with ColecoVision version of Donkey Kong. They may have even sought a partnership with Coleco to produce the U.S. Famicom. However, it was likely a combination of poor timing and sour grapes over the aborted Atari deal. Post-1983, no American video game company was ready to introduce a new console into the industry’s death spiral.

In the end, Nintendo had to forge ahead alone.

Postscript

I hate to leave this post open-ended, but there has been some discussion on the NESDev forums about the TMS9918’s influence on the Famicom design. The evidence I mentioned above is sourced from a mid-1990s article on Nintendo that is only available in Japanese.^[1] One thread commenter, 6502freak, writes that the article says Nintendo ‘first wanted to use the Z80, but chose the 6502 instead because it only takes 1/3 of the Z80 chip space.’

Perhaps that’s the simple hardware constraint that led them down a long and shady road to microchip espionage?

^[1] I have hired someone to translate the article, so as new information surfaces on the Coleco/Nintendo connection, I’ll be sure to post it here.