[SOLVED] PTAT error in BJT based temperature sensors?

[vaah]

Hi guys,

I have a doubt about trimming a temperature sensor. I have seen the fact that "for BJT based temperature sensors if an error/spread (for example current gain) is PTAT it can be easily trimmed out".
Can someone enlighten me? It kinda makes sense since the output signal of the temperature sensor is also PTAT but I need more clarification

Thanks.

 

[erikl]

The ease depends on what you consider as your final result: trimming of resistance values is mostly done by cutting any/several resistors from a chain of appropriate series or parallel resistors. Relatively easy to be done in schematics and simulation. In layout however the effort isn't quite so simple as cutting must be done at the top metal level, below openings in the passivation layer. And if this must be done for a high volume product, automatic trimming means measurement, calculation and (Laser or E-beam) cutting with a specialized equipment - not quite easy, so adds testing cost. Still, it's done for medium to high accuracy products.

 

[dick_freebird]

There are many methods of trim - laser is one that is
good for volume production of small analog parts in
wafer form. But this resource is not very common
any more outside of IDMs with captive test floors, and
maybe a few higher-end pure play test houses.

Fuse, zener-zap, EEPROM are other options, using
a shorting element across a regular resistor and
either opening or shorting electrically, the link. Of
course laser trim can also be done link-style and
often is, because open or shorted are both more
stable than the heat affected region abutting the
laser kerf. This can matter in MSB resistors etc.

Laser trim also needs the laser wavelength to
match the optical interference in the glass below
the trim material (destructive interference puts
too much power into everything else, etc.). The
NiCr / SiCr field oxide thickness is a process
control-point when laser trim is planned.

PTATs always have some curvature which in the
end limits the achievable accuracy even if trimming
is done (curvature correction in analog has been
a popular topic of papers, but its manufacturing
realities are not so great; trade a square law
problem for a cubic one, the sensitivities to some
process variations go up an order of magnitude
and may come to dominate yield but on more of
a "lot bust" scale.

In many fabs a fuse link is offered (along with the
pad cell design and test method info needed to
do it consistently. reliably and with a paper trail
that lets you qualify the design). Antifuse is less
common. Some technologies offer "all internal"
fuse (w/ on chip switches and serial programming)
but others have only the element and you're on
your own for delivering the programming energy
and sensing the trim element state.

Trimming might be "easy" (especially for some
paper author who resides in academia and has no
actual responsibility for outcomes). But behind
the capability, you need a technology that is
otherwise highly stable over life / stress (why
trim out a couple of mV of Vbe mismatch, when
you have 10mV of variable Vbe drift over burnin?).

 

[vaah]

There are many methods of trim - laser is one that is
good for volume production of small analog parts in
wafer form. But this resource is not very common
any more outside of IDMs with captive test floors, and
maybe a few higher-end pure play test houses.

Fuse, zener-zap, EEPROM are other options, using
a shorting element across a regular resistor and
either opening or shorting electrically, the link. Of
course laser trim can also be done link-style and
often is, because open or shorted are both more
stable than the heat affected region abutting the
laser kerf. This can matter in MSB resistors etc.

Laser trim also needs the laser wavelength to match the optical interference in the glass below
the trim material (destructive interference puts
too much power into everything else, etc.). The
NiCr / SiCr field oxide thickness is a process
control-point when laser trim is planned.

PTATs always have some curvature which in the
end limits the achievable accuracy even if trimming
is done (curvature correction in analog has been
a popular topic of papers, but its manufacturing
realities are not so great; trade a square law
problem for a cubic one, the sensitivities to some
process variations go up an order of magnitude
and may come to dominate yield but on more of
a "lot bust" scale.

In many fabs a fuse link is offered (along with the
pad cell design and test method info needed to
do it consistently. reliably and with a paper trail
that lets you qualify the design). Antifuse is less
common. Some technologies offer "all internal"
fuse (w/ on chip switches and serial programming)
but others have only the element and you're on
your own for delivering the programming energy
and sensing the trim element state.

Trimming might be "easy" (especially for some
paper author who resides in academia and has no
actual responsibility for outcomes). But behind
the capability, you need a technology that is
otherwise highly stable over life / stress (why
trim out a couple of mV of Vbe mismatch, when
you have 10mV of variable Vbe drift over burnin?).

Thank you very much @dick_freebird and @erikl .
From my understading a source of error need not neccesarily be a PTAT type to be trimmed out. A CTAT can be also trimmed, right?

Just to clarify, PTAT measn any signal that increaseing with tempearture while CTAT decreasing with temperature, is it true?

 

[erikl]

A CTAT can be also trimmed, right?

Sure!

Just to clarify, PTAT measn any signal that increaseing with tempearture while CTAT decreasing with temperature, is it true?

PTAT: proportional to absolute temperature
CTAT: complementary to absolute temperature, s. this Wiki explanation.