Ken's reverse engineering articles are always a real pleasure to read, but this one is special in that it looks at an analog chip and how typical analog building blocks (current mirrors, differential pairs) are implemented in silicon.
Also, if you've every looked at an analog IC's schematics and wondered what those weird "double collectors" BJT are (https://electronics.stackexchange.com/questions/105777/what-...) , there's a pretty decent description in there of what a BJT actually looks like when implemented in silicon.
Funny, I have actually used this one, the 4 means 4 OPAMPs in a single chip. IIRC there's also the TL082 with 2 OPAMPs.
Much more performant than the 741 (might have to do with 2 things: the 741 came early and was one of the pioneers and it is BJT only)
It's also curious how the the big butterfly transistors are at the input, components with a big die size are usually big for a reason (usually power).
One extra fact, the compensation capacitors are more like a wrench on gears and actually make the circuit "worse" (it lowers the overall bandwidth). But it is needed for stability purposes because good amplifiers will have a tendency to oscillate by themselves.
That may be true but I'm not sure it applies in this context. Another valid reason is because at a larger scale it is easier to get the dimensions to be within a fairly close tolerance of the counterpart and hence to get two parts that function well in tandem.
Matched pairs and current mirrors are exactly the right situations for this and so they tend to be oversized, even if that slows the parts down considerably.
If you build ~1MHz circuits with 741s and replace them with 071s, you'll notice how ass the 741s are.
Also, I've seen some people use 071s' inputs with no DC path to ground, and see their output slowly drift up/down because of tiny bias currents. Doesn't happen with 741s, usually.
I doubt there is anything recent as iconic. Overall I think we have moved way beyond having individual generic iconic models at all. Digikey has like thousands (hundreds if you are old-fashioned and want through-hole) of different opamps which probably are mostly interchangeable 741 for most purposes, just pick any one of them.
Also, there ARE objectively shit op amps that are old and crusty that pretty much nobody uses. I think it was the LM358 that basically had a broken output stage (class-B) that causes horrible output crossover distortion when you transition from sourcing to sinking current or vice versa.
Q12 is a common emitter amplifier with an active load, that feeds directly into a class B output stage (pullup is a darlington with Q5 and Q6, and pulldown is Q13, with Q7 being a current limiter). I've never used this personally (for good reason) but I remember my mentors telling me this causes horrible CO distortion.
I don't have any particular problem with it. I majored in EE about 15 years ago, and haven't really looked at op amps since studying the 741. Just curious what's popular these days.
While you marvel at this circuit remember that it was invented decades ago, for many purposes it is still state-of-the-art, and it costs fifteen cents.
I sometimes wonder whether integrated-circuit manufacturing wasn't sent to us by time-travellers from the future. Our ability to manufacture useful things, on such tiny scales, to such high precision, doesn't seem to match up with our comparatively-poor capabilities in other areas of manufacturing.
Many other things could be manufactured "on such tiny scales, to such high precision", but they wouldn't be useful.
Integrated circuits are one of a few product/technology classes for which miniaturization is unconditionally good (improved power consumption, lower unit cost, possibly higher speed) up to unavoidable physical limits (e.g. randomly arranged doping atoms, noise vulnerability).
Integrated circuits are also very easy to interface with the "real world" despite miniaturization because of their electronic nature: a comfortable range of currents and voltages is tolerable for the integrated circuit and acceptable for the larger system.
There's a series on transistors that covers this. There's basically no mechanical process since everything is done via etching and litho. Leads to the results you see.
- Crazy rail-to-rail input: some chips have onboard charge pumps to bias input stages such that you can input signals with common mode voltages far below and above the normal rails
- Rail-to-rail output: probably the lowest hanging fruit of the "new" features, but still pretty costly
- Stupidly high input impedance: 071 isn't completely blown out of the water but some of the new JFET opamps are nuts. I think there are like femptoamp class input currents these days.
- Stupidly low offset voltage: I'm not even talking about the auto-zero ones, just regular old op amps have amazing performance these days.
Not only invented decades ago, but built decades ago. The "7949" on the chip means it was manufactured in the 49th week of 1979. Which is nearly 39 years ago.
He's lucky to find a part like that. Today nobody is going to build a device in a 14-pin ceramic package and sell it for 15 cents.
I worked on a motor controller that had a problem with PWM IC frying.
Turned out the real problem was when the control signal was over-driven a comparator would latch up. Which would cause both power transistors to turn on. Blowing a 'protection' [1] fuse on the negative supply rail. And then the negative power supply would reverse. And kill the PWM controller IC.
Not a fan of latching OP-Amps and Comparators.
[1] Tip: fuses don't protect circuits from damage they prevent your POS from burning the house down.
Personally I prefer to have parts that require external protection, rather than parts with robust internal protections that cost more or compromise the performance or that I have to hack around. It's more work, board area, and overall cost to have the external protections to keep an IC within its safe operating area, but I don't mind.
I have a few questions about the JFET diagram and JFETs in general if someone wouldn't mind answering:
In the graphic we see S, G and D - Source Gate and Drain. Is the source permanently connected to power? If so do all transistor in a circuit connect to a shared power rail?
Must voltage always be present on both the source and gate in order for current to flow to the drain?
Lastly when the transistor is switched "on" does voltage leaving the drain then become input for a gate of some neighboring transistor in the circuit?
> Is the source permanently connected to power? If so do all transistor in a circuit connect to a shared power rail?
Looking at the TL084 schematic, the JFET source is fed from another transistor (a current source in a current mirror). Some of the transistors are connected to V++ and some are connected to V-- but many of them are not connected directly to any power rail.
> Must voltage always be present on both the source and gate in order for current to flow to the drain?
There has to be a voltage differential between the source and the drain for current to flow. JFETs are "normally on" and get pinched off as the voltage differential between the gate and source increases.
> Lastly when the transistor is switched "on" does voltage leaving the drain then become input for a gate of some neighboring transistor in the circuit?
Voltage doesn't really leave the drain, but the drain is connected to the base of another transistor, part of the second state of the op amp, where most of the amplification happens. In other words, the output of the differential pair is the input to the next amplifier stage.
Would these last two points also apply to MOSFETs?
In the case of a MOSFET there obviously wouldn't be a "next amplifier stage" but would the output voltage be input to the next transistor's source or the next transistor's gate maybe?
You can put these things one after the another to get more complex behavior.
Roughly speaking, common collector/common drain amplifiers have a gain of approximately 1, but has low output impedance, which makes it suitable for driving big loads (like the outside world). Common emitter/common source amplifiers have high output impedance but tons of gain.
The classic op-amp topology is a differential amplifier (long-tailed pair, in the wikipedia template) feeding into a common emitter/source gain stage, then into a common collector/drain follower output stage.
To answer your last question, the output of the gain stage is the drain of the transistor, that feeds the gate of the common drain output stage transistor.
The gate is usually the input terminal, but sometimes the gate is at a fixed voltage and the drain is the input terminal (cascode-y circuits are like this, aka common base or common gate).
Does anyone have a good reference on how analog is done on logic processes, specifically the bits that have to interface off the chip, like gpio, lvds, and serdes. I'd much appreciate it.
There's something I once encountered, but can't find the reference to, that suggested a kind of "digital-analog" process whereby voltage levels were replaced by timing measurement (?) due to the limits of feature size in analog design. I'm not in the field so forgive my ignorance.
I don't know about modern chips, but 1970s and 1980s logic chips just used big MOSFETs to drive their outputs. There wasn't any weird timing magic going on.
There are also processes like analog BiCMOS that let you mix bipolar analog circuitry and CMOS on the same chip. I also wrote recently about the 76477 sound chip that combined I2L logic with bipolar analog circuits.
Thanks for pointing out your article about the 76477. Fascinating stuff.
As for my second paragraph about "digital-analog", this reference on VCO-based quantizers [1] is not the one I remember but I think is related. Not that I really understand the intricacies, but it's a way of doing analog at low-voltages and finer geometries.
I'd be interested as well; even when I was working with digital logic designers they described this process (eg bandgap references and PLLs) as "black magic", and it's very rare in undergrad or even master's courses.
Not my area but I don't think that's the case for the analog ICs, or the analog part of mixed ICs.
The constraints for generic digital logic soup are comparatively "simple" : make wires as short and neat as possible and then check the (simulated) physical timing characteristics.
Analog is more artistic since any noise or crosstalk degrades the signal irreversibly (for low noise stuff) and any wire is a transmission line (for high speed stuff).
Actually even for digital I'm sure you still have to do manual layout for the most critical pieces of high-performance designs (say a register file on a nvidia gpu).
Also, if you've every looked at an analog IC's schematics and wondered what those weird "double collectors" BJT are (https://electronics.stackexchange.com/questions/105777/what-...) , there's a pretty decent description in there of what a BJT actually looks like when implemented in silicon.
Much more performant than the 741 (might have to do with 2 things: the 741 came early and was one of the pioneers and it is BJT only)
It's also curious how the the big butterfly transistors are at the input, components with a big die size are usually big for a reason (usually power).
One extra fact, the compensation capacitors are more like a wrench on gears and actually make the circuit "worse" (it lowers the overall bandwidth). But it is needed for stability purposes because good amplifiers will have a tendency to oscillate by themselves.
That may be true but I'm not sure it applies in this context. Another valid reason is because at a larger scale it is easier to get the dimensions to be within a fairly close tolerance of the counterpart and hence to get two parts that function well in tandem.
Matched pairs and current mirrors are exactly the right situations for this and so they tend to be oversized, even if that slows the parts down considerably.
Also, I've seen some people use 071s' inputs with no DC path to ground, and see their output slowly drift up/down because of tiny bias currents. Doesn't happen with 741s, usually.
http://www.ti.com/lit/ds/symlink/lm158-n.pdf
Page 13, Figure 16.
Q12 is a common emitter amplifier with an active load, that feeds directly into a class B output stage (pullup is a darlington with Q5 and Q6, and pulldown is Q13, with Q7 being a current limiter). I've never used this personally (for good reason) but I remember my mentors telling me this causes horrible CO distortion.
The 071 is also "jellybean" enough; both are still currently used in modern designs because how dirt cheap they are.
But they are "high voltage", non rail-to-rail anything, so people tend to use more modern options for new designs.
They’re just horrid. Fine for noddy stuff like power supplies and PID loops etc though.
Integrated circuits are one of a few product/technology classes for which miniaturization is unconditionally good (improved power consumption, lower unit cost, possibly higher speed) up to unavoidable physical limits (e.g. randomly arranged doping atoms, noise vulnerability). Integrated circuits are also very easy to interface with the "real world" despite miniaturization because of their electronic nature: a comfortable range of currents and voltages is tolerable for the integrated circuit and acceptable for the larger system.
[citation needed]
"New" op amp features:
- Crazy rail-to-rail input: some chips have onboard charge pumps to bias input stages such that you can input signals with common mode voltages far below and above the normal rails
- Rail-to-rail output: probably the lowest hanging fruit of the "new" features, but still pretty costly
- Stupidly high input impedance: 071 isn't completely blown out of the water but some of the new JFET opamps are nuts. I think there are like femptoamp class input currents these days.
- Stupidly low offset voltage: I'm not even talking about the auto-zero ones, just regular old op amps have amazing performance these days.
He's lucky to find a part like that. Today nobody is going to build a device in a 14-pin ceramic package and sell it for 15 cents.
Turned out the real problem was when the control signal was over-driven a comparator would latch up. Which would cause both power transistors to turn on. Blowing a 'protection' [1] fuse on the negative supply rail. And then the negative power supply would reverse. And kill the PWM controller IC.
Not a fan of latching OP-Amps and Comparators.
[1] Tip: fuses don't protect circuits from damage they prevent your POS from burning the house down.
In the graphic we see S, G and D - Source Gate and Drain. Is the source permanently connected to power? If so do all transistor in a circuit connect to a shared power rail?
Must voltage always be present on both the source and gate in order for current to flow to the drain?
Lastly when the transistor is switched "on" does voltage leaving the drain then become input for a gate of some neighboring transistor in the circuit?
Looking at the TL084 schematic, the JFET source is fed from another transistor (a current source in a current mirror). Some of the transistors are connected to V++ and some are connected to V-- but many of them are not connected directly to any power rail.
> Must voltage always be present on both the source and gate in order for current to flow to the drain?
There has to be a voltage differential between the source and the drain for current to flow. JFETs are "normally on" and get pinched off as the voltage differential between the gate and source increases.
> Lastly when the transistor is switched "on" does voltage leaving the drain then become input for a gate of some neighboring transistor in the circuit?
Voltage doesn't really leave the drain, but the drain is connected to the base of another transistor, part of the second state of the op amp, where most of the amplification happens. In other words, the output of the differential pair is the input to the next amplifier stage.
Would these last two points also apply to MOSFETs?
In the case of a MOSFET there obviously wouldn't be a "next amplifier stage" but would the output voltage be input to the next transistor's source or the next transistor's gate maybe?
For the tree transistor horsemen (BJTs, MOSFETs, JFETs), they can all be fitted into similar transistor amplifier building blocks:
https://en.wikipedia.org/wiki/Template:Transistor_amplifiers
You can put these things one after the another to get more complex behavior.
Roughly speaking, common collector/common drain amplifiers have a gain of approximately 1, but has low output impedance, which makes it suitable for driving big loads (like the outside world). Common emitter/common source amplifiers have high output impedance but tons of gain.
The classic op-amp topology is a differential amplifier (long-tailed pair, in the wikipedia template) feeding into a common emitter/source gain stage, then into a common collector/drain follower output stage.
To answer your last question, the output of the gain stage is the drain of the transistor, that feeds the gate of the common drain output stage transistor.
The gate is usually the input terminal, but sometimes the gate is at a fixed voltage and the drain is the input terminal (cascode-y circuits are like this, aka common base or common gate).
There's something I once encountered, but can't find the reference to, that suggested a kind of "digital-analog" process whereby voltage levels were replaced by timing measurement (?) due to the limits of feature size in analog design. I'm not in the field so forgive my ignorance.
There are also processes like analog BiCMOS that let you mix bipolar analog circuitry and CMOS on the same chip. I also wrote recently about the 76477 sound chip that combined I2L logic with bipolar analog circuits.
As for my second paragraph about "digital-analog", this reference on VCO-based quantizers [1] is not the one I remember but I think is related. Not that I really understand the intricacies, but it's a way of doing analog at low-voltages and finer geometries.
[1] http://ewh.ieee.org/r5/central_texas/cas_ssc/meetings/2012/1...
(Electronics seems an interesting application area for automated design, since the search space is relatively small)
The constraints for generic digital logic soup are comparatively "simple" : make wires as short and neat as possible and then check the (simulated) physical timing characteristics.
Analog is more artistic since any noise or crosstalk degrades the signal irreversibly (for low noise stuff) and any wire is a transmission line (for high speed stuff).
Actually even for digital I'm sure you still have to do manual layout for the most critical pieces of high-performance designs (say a register file on a nvidia gpu).