4
\$\begingroup\$

Historically, how did the density of bipolar (e.g. TTL) chips compare with MOS? How complex could bipolar ICs get before they hit limits? Does anyone have a Moore's law-style graph for bipolar technology?

Edit: A specific number I'd like to know is the most transistors or gates implemented on a bipolar IC.

Background, and what I've found so far:

I'm looking at the TTL 74181 ALU chip (1970, 75 logic gates), which was a fairly advanced bipolar chip for its time. At the time, MOS was somewhat ahead in density with thousands of transistors on a chip. But now MOS ICs can have billions of transistors but bipolar is nowhere near that. That got me wondering at what point did TTL (or other bipolar technologies) stop scaling?

I came across Integrated Injection Logic (I2L), a bipolar technology that was supposed to provide MOS-like density. The TI SPB9900 microprocessor (1976) used I2L and had 6034 gates, which surprisingly is more than MOS microprocessors from 1976. I haven't heard much about I2L, so did it die out?

In 1986 there was the MBM10494 64K bipolar ECL RAM, versus 256K MOS DRAM, so bipolar was still fairly close. I haven't been able to find larger bipolar RAMs, so I suspect that's about the limit. Was the limiting factor power consumption, manufacturing, economics, or something else?

\$\endgroup\$
  • \$\begingroup\$ Moore's law has nothing to say about a particular technology. What you're asking for is closer to Darwinian evolution - survival of the fittest (technology) which is still going on and has been for a long time. We used valves (tubes) before we discovered the transistor. The transistor got faster. We discovered FETs. We used Silicon instead of Germanium. We made simple ICs with several transistors. We learned to produce more complex circuits and so on. What about valves (tubes) today? - they're still around but consigned to 'a curiosity corner', like the other obsolete stuff - just in case! \$\endgroup\$ – JIm Dearden Jan 3 '17 at 19:36
  • 1
    \$\begingroup\$ If you want to know what lies at the cutting edge, I think e2v uses a SiGe bipolar process and they've got things like a quad channel Gsps class ADC. But as was Jim alluding too, MOS took over from bipolar not necessarily because it scaled better, but because it was WAY more power efficient. \$\endgroup\$ – Sam Jan 3 '17 at 20:58
  • \$\begingroup\$ I wonder how much power would an ARM MCU made out of bipolar (if it was possible) consume compared to the same design made with MOS. Seriously, bipolar is, of course, still useful in a lot of applications. But putting millions of them on a chip would certainly make it unuseable, so I understand that research engineers prefer to spend their time on MOS technology. \$\endgroup\$ – dim Jan 3 '17 at 21:18
  • 1
    \$\begingroup\$ Signetics made 8X300 and 8X305 CPUs in Schottky TTL, all bipolar. \$\endgroup\$ – Whit3rd Jan 8 '17 at 11:43
  • \$\begingroup\$ ECL was particularly power hungry (25mW per gate), although at the time it was the fastest logic on the market. Those speeds have long since been eclipsed by low power CMOS. \$\endgroup\$ – Peter Smith Jan 8 '17 at 15:07
1
\$\begingroup\$

CMOS eclipsed bipolar technologies because of the power savings. All bipolar logic draws substantial current when idle and, since at least the 1990s, it becomes impossible to get the heat out of the chip long before you run out of space for transistors.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.