Op-amp design: Bipolar or Cmos

[linnix]

Yes, we need diodes in there too.

Just buy one of the shelf it will be millions cheaper !

Not when they want it as small and light weight as possible. The
chip, board and case sizes and weights will all add-up. The customer
is serious about paying $30,000 more (out of $100,000) in NRE, but I
have to wait and see if they will write the check. Meanwhile, they
are still paying for other stuffs, so I have to proceed as planned.

I am just nervous about mistakes. PCB errors are hundred dollars.
Wafer errors are thousand dollars. We will have multiple designs in
the chip and activate some with one metal layer. Inactive designs
will stay there, since it cost more (masks) to remove them. It's kind
of a OTP analog gate array.

By the way, the bipolar design needs 8 masks. Any idea on a cmos
design? Masks are the most expensive NRE.

 

[Joerg]

I met Widlar's girlfriend at the bar at the Washingtin Square Bar and
Grille, in San Francisco. They were living in Puerto Vallarta at the
time, and she was back in town visiting family. She was surprised that
I'd heard of him... she had no idea how famous he was. She saad that
he kept his designs under the kitchen tablecloth and she was not
allowed to look. He only worked on them when she was out.

She was concerned that he was drinking too much.

He must have been a true engineer to the bones. They typicaly do not
brag about their famousness level. Sad that he isn't with us anymore.

 

[Spehro Pefhany]

LM358 is cool: a decent 741-type dual opamp in SO-8 for around 12
cents! That's 6 cents per opamp!

It's a bit irritating that it costs about the same as the LM324 but
has half the number of the same kind of op-amps.

In my youth, my mentor Melvin Goldstein told me that, some day,
transistors inside ICs would cost a penny each. I thought he was nuts.

John

Best regards,
Spehro Pefhany

 

[Spehro Pefhany]

I don't want to start a flame war (probably did). The best IC
designers I know are C programmers, who wrote the design programs.

Of course!

And the very best authors are surely those folks who wrote the word
processing programs.



Best regards,
Spehro Pefhany

 

[Joerg]

Chris Jones wrote:


[...]

To throw a further spanner in the works, you may well find that an opamp is
not the optimal kind of amplifier to do what you need to do. For example,
it is difficult to buy stand-alone OTAs, but if you are designing your own
chip then you are free from these restrictions and you might find that
there are things that will achieve your goals better than an op-amp would.

Exactly. Also, one should keep in mind that simple amplification jobs
can be done with the same scheme that digital inverters are made of. In
contrast to us "discreteros" the chip guys have the luxury of picking a
device geometry and thus a reasonable cross current for a desired VCC.
We have to do it the other way around, adjust VCC so the cross current
is reasonable.

Come to think of it, during my chip designs I never had the urge for an
opamp, ever. We always did things directly, whether it was comparing two
inputs, amplification, multiplying, whatever. There was a nice simple
chip solution for just about any local job and it was a lot smaller than
a full opamp.

 

[Chris Jones]

We are in the process of building some op-amp ICs.
I found a reference design and layout of a bipolar op amp.
However, the author says that cmos op-amps are more popular
in the real world. Questions:

#1 Anybody got a cmos op-amp design and layout to sell or license?

#2 Anybody able to port from bipolar to cmos?

This will not give you an optimal design. If you use mosfets like you would
want to use bipolars, you will find that they make lousy bipolars. If you
accept that they are mosfets, and design your opamp from scratch as a CMOS
opamp, that will work much better.

#3 What are the drawbacks to stay with bipolar, other than powers?

If you have more than a few hundred logic gates in the other part of your
chip, and you want low power consumption or a small die, CMOS or BiCMOS
probably is better than bipolar. (BiCMOS = bipolar + CMOS, best of both
worlds and most expensive process too) If you have tens of thousands of
logic gates on your die then you definitely want either CMOS or BiCMOS.
You could still use a bipolar opamp if you use a BiCMOS process.

The op-amp circuit will be enable on demand, so power usages
may not be too critical for stand-by. What about active power usages
between bipolar and cmos?

What are your specs? Otherwise how long is a piece of string?

How about linearity? Would bipolar be better than cmos?

You can do well in either technology with sufficient effort.

Unless the opamp is the principal function of your chip (which would make it
a very strange business proposition), the rest of the chip will probably
determine the technology that you need to use. First of all, what is the
rest of the chip? How many gates of digital (if any), are you designing it
yourself or buying IP blocks, what is the max supply voltage you need, what
kind of frequencies does the chip need to handle and how does it need to
handle them, and what production volume do you anticipate? How much can
you afford to spend on masks?

Also you will need to list the specs of your amplifier - offset voltage,
bias current, do you need the inputs to work near both supply rails or only
near one supply rail (rail to rail inputs can be a nuisance and should be
avoided if you want the best linearity performance without getting into
patent difficulties), do you need the output to go near both rails or only
one, how many milliamps output current do you need, etc. etc. Without
answering these questions and many more, it will be a case of garbage in ->
garbage out.

To throw a further spanner in the works, you may well find that an opamp is
not the optimal kind of amplifier to do what you need to do. For example,
it is difficult to buy stand-alone OTAs, but if you are designing your own
chip then you are free from these restrictions and you might find that
there are things that will achieve your goals better than an op-amp would.

Chris

 

[Tam/WB2TT]

And certainly not a CLUE ;-)

...Jim Thompson

I looked that up in 2 dictionaries, one had CLUE, see CLEW, the other had
CLEW, alternate spelling for CLUE. Take your pick. BTW, my point about the
managers with non engineering backgrounds is that they will, for instance,
take a logic circuit designer, and assign him to design a microwave
amplifier or chip, and not understand why it takes the guy a while to get
on board.

Tam

 

[Jim Thompson]

[snip]

This comes about because management consists of MBAs and C programmers,
neither of which has a clew about engineering.

Tam

And certainly not a CLUE ;-)

...Jim Thompson

I looked that up in 2 dictionaries, one had CLUE, see CLEW, the other had
CLEW, alternate spelling for CLUE. Take your pick. BTW, my point about the
managers with non engineering backgrounds is that they will, for instance,
take a logic circuit designer, and assign him to design a microwave
amplifier or chip, and not understand why it takes the guy a while to get
on board.

Tam

Sure it wasn't a _girolle_ ?

...Jim Thompson

 

[John Devereux]

I looked that up in 2 dictionaries, one had CLUE, see CLEW, the other had
CLEW, alternate spelling for CLUE. Take your pick.

Methinks it Soundes like ye Olde English speling!

 

[linnix]

This will not give you an optimal design. If you use mosfets like you would
want to use bipolars, you will find that they make lousy bipolars. If you
accept that they are mosfets, and design your opamp from scratch as a CMOS
opamp, that will work much better.


If you have more than a few hundred logic gates in the other part of your
chip, and you want low power consumption or a small die, CMOS

I know it's a strange requirement, but size is not an issue.
Basically, the analog circuits (op-amp, charge pump, etc) will be
added to an area of 2000 to 3000 microns rectangle. The area would
otherwise be wasted anyway. There are no other active circuit in this
wafer, which will be combined with another wafer before dicing.

or BiCMOS
probably is better than bipolar. (BiCMOS = bipolar + CMOS, best of both
worlds and most expensive process too)

High mask count is out of the question. We would rather pick one with
least mask set.

If you have tens of thousands of
logic gates on your die then you definitely want either CMOS or BiCMOS.
You could still use a bipolar opamp if you use a BiCMOS process.


What are your specs? Otherwise how long is a piece of string?

We don't know exactly until the final device is tested.

You can do well in either technology with sufficient effort.

Unless the opamp is the principal function of your chip (which would make it
a very strange business proposition),

The opamp would be the first to be integrated. Other parts will just
follow with the same process, hopefully.

the rest of the chip will probably
determine the technology that you need to use. First of all, what is the
rest of the chip? How many gates of digital (if any), are you designing it
yourself or buying IP blocks, what is the max supply voltage you need,
18V

what kind of frequencies
10KHz

does the chip need to handle and how does it need to
handle them, and what production volume do you anticipate?
100,000+

How much can you afford to spend on masks?

Less than 10 additional masks.

Also you will need to list the specs of your amplifier - offset voltage,
bias current, do you need the inputs to work near both supply rails or only
near one supply rail (rail to rail inputs can be a nuisance and should be
avoided if you want the best linearity performance without getting into
patent difficulties),

No, we can charge pump the supply high enough to avoid rail to rail
requirements.

do you need the output to go near both rails or only
one, how many milliamps output current do you need, etc. etc.

Just enough to be A2D by an uC.

Without
answering these questions and many more, it will be a case of garbage in ->
garbage out.

To throw a further spanner in the works, you may well find that an opamp is
not the optimal kind of amplifier to do what you need to do. For example,
it is difficult to buy stand-alone OTAs, but if you are designing your own
chip then you are free from these restrictions and you might find that
there are things that will achieve your goals better than an op-amp would.

We know the structure of the design, but not the exact parameters at
this point. However, we can redo a couple of masks in the final
iteration. For example, adjusting the resistive values with the metal
layer.

 

[David DiGiacomo]

He must have been a true engineer to the bones. They typicaly do not
brag about their famousness level. Sad that he isn't with us anymore.

Huh? Widlar had a huge ego and knew exactly how famous he was. His
girlfriend probably just didn't believe his bragging.

Anyway, this thread is giving me a headache. Way too much vagueness.

Hey OP, you are building some kind of chip, right? But then you mentioned
a wafer cap, so it's MEMS??!? Why would you put the op amp on the wafer cap
instead of the base chip?

Anyway, you must have a foundry... so cut out the guesswork and ask them
what op amp IP they have. They can tell you what process makes sense also.

 

[linnix]

Huh? Widlar had a huge ego and knew exactly how famous he was. His
girlfriend probably just didn't believe his bragging.

Anyway, this thread is giving me a headache. Way too much vagueness.

Hey OP, you are building some kind of chip, right? But then you mentioned
a wafer cap, so it's MEMS??!?

Yes, the MEMS takes up all the space of the lower wafer.

Why would you put the op amp on the wafer cap
instead of the base chip?

They are two different processes and no space on the base anyway.

Anyway, you must have a foundry... so cut out the guesswork and ask them
what op amp IP they have. They can tell you what process makes sense also.

They could be from different fab houses as well.

 

[Jim Thompson]

I know it's a strange requirement, but size is not an issue.
Basically, the analog circuits (op-amp, charge pump, etc) will be
added to an area of 2000 to 3000 microns rectangle. The area would
otherwise be wasted anyway. There are no other active circuit in this
wafer, which will be combined with another wafer before dicing.


High mask count is out of the question. We would rather pick one with
least mask set.


We don't know exactly until the final device is tested.


The opamp would be the first to be integrated. Other parts will just
follow with the same process, hopefully.


Less than 10 additional masks.


No, we can charge pump the supply high enough to avoid rail to rail
requirements.


Just enough to be A2D by an uC.


We know the structure of the design, but not the exact parameters at
this point. However, we can redo a couple of masks in the final
iteration. For example, adjusting the resistive values with the metal
layer.

Are you saying you need an 18V OpAmp? That arrows you down to just a
couple of foundries in the WORLD. As you've already been advised...
get thee hence to a foundry.

...Jim Thompson

 

[Joerg]

Methinks it Soundes like ye Olde English speling!

Then I guess it would be "soundeth", methinks

 

[linnix]

Are you saying you need an 18V OpAmp?

Probably 12V to 15V. 3V to 5V is not enough.

That arrows you down to just a ^ keyboard check ahead
couple of foundries in the WORLD.

I know. We have to look for oldest technologies in certain area, but
the newest in others. The top and bottom wafers are totally different
technologies. We are also checking out different bonding and dicing
processes.

 

[Joerg]

Probably 12V to 15V. 3V to 5V is not enough.



^ keyboard check ahead



I know. We have to look for oldest technologies in certain area, but
the newest in others. The top and bottom wafers are totally different
technologies. We are also checking out different bonding and dicing
processes.

And be careful when trying this stuff on a shoestring budget. BTDT (5um
precision flip chip bonding and stuff like that...).

 

[Jim Thompson]

[snip]

Are you saying you need an 18V OpAmp?

Probably 12V to 15V. 3V to 5V is not enough.

That arrows you down to just a

^ keyboard check ahead

My new keyboard has stiff keys :-(

I know. We have to look for oldest technologies in certain area, but
the newest in others. The top and bottom wafers are totally different
technologies. We are also checking out different bonding and dicing
processes.

...Jim Thompson

 

[linnix]

And be careful when trying this stuff on a shoestring budget. BTDT (5um
precision flip chip bonding and stuff like that...).

Actually, we can make the pads very big, even 10s of um. The
difficult part is going through the top layer from the bottom. We
would need to etch holes on the top wafer and metal fill them, and
deposit metal on both sides. It sounds simple. But whenever we ask
what the fab can do, they will say they can do whatever we ask
(subject to their DRC).

We are in a funny position that the bottom wafer is very big and the
top wafer is very sparse. So, we want to fill the top wafer with
different circuits, just to share the fab costs. The op-amp building
blocks and charge pumps could be built in the same wafer, but be
packed in different chips. We definitely need a safe distance between
them.

Jim is right, the voltage requirement would likely drive us into
bipolar anyway, at least on one side of the top wafer.

 

[Joerg]

Actually, we can make the pads very big, even 10s of um. The
difficult part is going through the top layer from the bottom. We
would need to etch holes on the top wafer and metal fill them, and
deposit metal on both sides. It sounds simple. But whenever we ask
what the fab can do, they will say they can do whatever we ask
(subject to their DRC).

Even DRCs can go out the window. Twice we absolutely had to because of
size constraints. The foundry just made us sign a waiver, basically
stating "we told ya that it ain't kosher but ya want it your way
anyhow". It was successfully built that way for more than a decade.

We are in a funny position that the bottom wafer is very big and the
top wafer is very sparse. So, we want to fill the top wafer with
different circuits, just to share the fab costs. The op-amp building
blocks and charge pumps could be built in the same wafer, but be
packed in different chips. We definitely need a safe distance between
them.

Can't you all sign your names up there? SCNR...

Jim is right, the voltage requirement would likely drive us into
bipolar anyway, at least on one side of the top wafer.

Probably. Although, 18V doesn't sound too harsh. After all, the CD4000
series is rated up there. There are also some HV processes but your
foundry might not offer any. We had the luxury of picking from foundries
like ABB Hafo and automotive.

 

[Rene Tschaggelar]

Are you saying you need an 18V OpAmp? That arrows you down to just a
couple of foundries in the WORLD. As you've already been advised...
get thee hence to a foundry.

...Jim Thompson

Jim,
You are saying CMOS OpAmps are limited to supply voltages
below 18V ? Amazing. I wasn't aware of that. Standard
bipolar OpAmps extend to the double.

Rene