3 ms·
So many companies had their own custom chips back then. Nowadays even game consoles use parts that are off the shelf PC components. Was it cheaper to make cust
by matheusmoreira 3y ago
So many companies had their own custom chips back then. Nowadays even game consoles use parts that are off the shelf PC components.
Was it cheaper to make custom hardware in that era?
- bantunes 3y agoI'd say not only cheaper, but the industry hadn't yet standardized regarding graphics technology so a lot of companies were trying different things (and not selling them to competitors at all).
- jonny_eh 3y agoNot only cheaper, but local. Many chips were still made in Japan or US, depending on where you needed them. To be fair, making modern chips is a lot more challenging, so it makes sense that it has become consolidated somewhat.
- Cthulhu_ 3y agoModern chips sure, but what about chips from a few generations ago? Do the factories they made 20 years ago for what was then cutting-edge tech still exist and produce things using that era of chip making tech, or have they been modernized?
- rwmj 3y agoVery much so. For example the UK's only FAB in Newport was formerly used to make Inmos Transputers. It can currently do 180nm chips which is late 90s technology. There are very many in Japan if you look at Wikipedia's list: https://en.wikipedia.org/wiki/List_of_semiconductor_fabrication_plants https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...
- rwmj 3y agoDon't know why I wrote "FAB" there. Maybe channelling Thunderbirds ...
- pjc50 3y ago> Do the factories they made 20 years ago for what was then cutting-edge tech still exist They're phasing out gradually. I work for an IC company and we're retooling some mostly-analogue parts from 10+ years ago onto modern processes because the right sort of fab is increasingly hard to find. I suspect that even if you dig an old GDSII file out from the nineties for one of those chips it may be hard to find somewhere to make it without requiring tweaks.
- mips_r4300i 3y agoAlmost all of the "custom" chips on these arcade systems through the 90s were not fully custom, they were gate arrays, sort of a fixed form of FPGA. You'd select from a few templates depending on how big a die you needed, or any distribution of special features like sram blocks, etc. And on this die would just be a giant grid or sea of gates, not wired together besides power and clocking. Then you would use the vendors tools to take your logic (in schematic format or very early HDL like ABEL) and they would generate a routing solution using 2 metal layers of interconnects to wire all the gates together. Then you could order a few tens of thousands of chips, and the fab would put your 2 metal layers on top of the silicon and package it. The cost was honestly not bad at all. NEC's prices ran about $50k of tooling (not including any software or design tool rental) which is peanuts. The per unit cost was higher than fully custom silicon though. Companies offering gate arrays were NEC, Toshiba, Ricoh, Fujitsu, Sharp, and so on. By the early 2000s this started to wane and NEC started pushing more incentives to keep the business going but now they much less common. There were services that would take a working FPGA design and handle the conversion for you. Gate arrays were a stepping stone in the journey of integration and consolidation. If you want to read more, look up a NEC Semiconductor Solutions Databook from the 90s.
- MrBuddyCasino 3y agoThanks, I always wondered how these chips were economical. I have no knowledge of chip design, did I understand correctly that there were a few pre-made "templates" chips populated by generic gate arrays (and supposedly SRAM and general purpose peripherals similar to MCUs?), you'd choose the one that best fit your needs, and then the existing gates of that template would be connected according to your custom logic needs using a "slapped on" metal layer?
- pjc50 3y agoSort of - the older custom hardware was smaller and simpler, so it could be done by a smaller team with simpler tools (or by hand with rubylith like the 6502!), and chips had comparatively cheaper processes with fewer layers. So the break-even point for "custom vs adapt non-custom" was at a smaller number of units. It could also be a source of real competitive advantage.