diode-connected transistors

[Terry Given]

I often use diode-connected npn and pnp transistors. Sometimes I use the
base-emitter junction reverse-breakdown as a (somewhat dodgy) zener,
sometimes I use them as diodes. When used as a diode, I can either use
the c-b junction (b shorted to e) or the b-e junction (b shorted to c).

Here are my questions:
1. Which configuration has the lowest capacitance ?
2. which configuration has the lowest leakage current?

In a more general sense, what are the pros and cons of either
connection? Using say bc547 and bc557 transistors.

Cheers
Terry

 

[Jim Thompson]

I often use diode-connected npn and pnp transistors. Sometimes I use the
base-emitter junction reverse-breakdown as a (somewhat dodgy) zener,
sometimes I use them as diodes. When used as a diode, I can either use
the c-b junction (b shorted to e) or the b-e junction (b shorted to c).

Here are my questions:
1. Which configuration has the lowest capacitance ?
2. which configuration has the lowest leakage current?

In a more general sense, what are the pros and cons of either
connection? Using say bc547 and bc557 transistors.

Cheers
Terry

CJE is generally larger than CJC (see below), but not necessarily so,
since capacitance is a function of both doping level _and_ area.

A B-E junction has lower breakdown than B-C.

..MODEL QBC547A/PLP_XN NPN(IS=15.33f NF=1.002 ISE=0.7932f NE=1.436
BF=178.7 IKF=121.6m VAF=69.7 NR=1.004
+ ISC=83.05f NC=1.207 BR=8.628 IKR=112.1m VAR=44.7 RB=1 IRB=1u RBM=1
+ RE=639.5m RC=650.8m XTB=0 EG=1.11 XTI=3 CJE=16.1p VJE=420.9m
MJE=307.1m
+ TF=499.5p XTF=139 VTF=3.523 ITF=702.1m PTF=0 CJC=4.388p VJC=200m
MJC=279.3m
+ XCJC=619.3m TR=1.e-032 CJS=0 VJS=750m MJS=333m FC=776.2m)

..MODEL QBC557A/PLP_XN PNP(IS=20.59f NF=1.003 ISE=2.971f NE=1.316
BF=227.3 IKF=87.19m VAF=37.2 NR=1.007
+ ISC=13.39f NC=1.15 BR=7.69 IKR=76.46m VAR=11.42 RB=1 IRB=1u RBM=1
+ RE=688m RC=643.7m XTB=0 EG=1.11 XTI=3 CJE=14p VJE=591.2m MJE=357.2m
+ TF=704.6p XTF=4.217 VTF=5.367 ITF=194.7m PTF=0 CJC=11.13p VJC=100m
MJC=341.4m
+ XCJC=628.8m TR=1.e-032 CJS=0 VJS=750m MJS=333m FC=794.7m)

...Jim Thompson

 

[Dingus]

Diodes are cheaper, more predictable, accurate and reliable than cobbling up
some home made diode-like device.

 

[John Larkin]

I often use diode-connected npn and pnp transistors. Sometimes I use the
base-emitter junction reverse-breakdown as a (somewhat dodgy) zener,
sometimes I use them as diodes. When used as a diode, I can either use
the c-b junction (b shorted to e) or the b-e junction (b shorted to c).

Here are my questions:
1. Which configuration has the lowest capacitance ?
2. which configuration has the lowest leakage current?

In a more general sense, what are the pros and cons of either
connection? Using say bc547 and bc557 transistors.

Cheers
Terry

Some NPNs make superb reference zeners if operated as a zener, C-E
with the base open. The tc of the forward-biased cb junction cancels
the tc of the zenering be junction; this works right at about 6.2
volts overall.

John

 

[Pooh Bear]

I often use diode-connected npn and pnp transistors. Sometimes I use the
base-emitter junction reverse-breakdown as a (somewhat dodgy) zener,

I believe at's actually avalanche breakdown.

sometimes I use them as diodes. When used as a diode, I can either use
the c-b junction (b shorted to e) or the b-e junction (b shorted to c).

Here are my questions:
1. Which configuration has the lowest capacitance ?
2. which configuration has the lowest leakage current?

In a more general sense, what are the pros and cons of either
connection? Using say bc547 and bc557 transistors.

Can't think of a good reason to do it to be honest, although I've used the
b-e junction to measure temperature ( because TO-220 devices are easy to
mount in a thermally meaningful way ).

Graham

 

[John Larkin]

Can't think of a good reason to do it to be honest, although I've used the
b-e junction to measure temperature ( because TO-220 devices are easy to
mount in a thermally meaningful way ).

There is a TO-220 version of the LM35. Handy on heat sinks.

John

 

[Terry Given]

Diodes are cheaper, more predictable, accurate and reliable than cobbling up
some home made diode-like device.

that all depends now, doesnt it. If you are designing with smt,
minimising part types is extremely important. There are also often
economies of scale. one of my customers uses a LOT (millions) of
BC847bpn dual npn/pnp transistors in a little 6-legged package. If I
need a few diodes it is a LOT cheaper to use one of the existing
transistors, than add a whole new reel. Some of my other customers use a
few of these parts - and converting a bunch of diodes into
diode-connected transistors saves a slot on the smt machine (and for a
contract manufacturer that's one less slot to screw up) and can
significantly increase the volume of the one part purchased, thereby
leading to further cost reductions.

Economic considerations aside, there is nothing at all "wrong" with
using a diode-connected transistor. pick an analogue IC, any analogue IC
and look inside. Hell, any decent discrete audio amp likely has them in
the current mirrors. Go read Grey & Meyer.....

Although b-e breakdown voltage is not a parameter to rely on, especially
between manufacturers. But nevertheless can still be useful.

Cheers
Terry

 

[Terry Given]

Some NPNs make superb reference zeners if operated as a zener, C-E
with the base open. The tc of the forward-biased cb junction cancels
the tc of the zenering be junction; this works right at about 6.2
volts overall.

John

Hi John,

Thanks, I'll try that. Care to elaborate on "some" ?

Thanks,
Terry

 

[Terry Given]

Terry Given wrote:




I believe at's actually avalanche breakdown.

odds on. Hell, any "zener" over what, 5V or so (too lazy to look it up)
is actually an avalanche diode, but we still call them zeners. Think how
tedious typing "zener (avalanche) diode" everywhere you use "zener"
would become

Can't think of a good reason to do it to be honest, although I've used the
b-e junction to measure temperature ( because TO-220 devices are easy to
mount in a thermally meaningful way ).

Graham

see above reply to dingus for the reasons.

I've forward-biased the input ESD diode in a variety of semiconductors
(micro, CPLD etc) to measure Tj.

Cheers
Terry

 

[Terry Given]

Hi Jim,

CJE is generally larger than CJC (see below), but not necessarily so,
since capacitance is a function of both doping level _and_ area.

So usually the lowest capacitance diode is the B-C junction, ie B-E
shorted, but as usual *it depends* on the actual device used

A B-E junction has lower breakdown than B-C.

I presume the B-C junction will break down at Vcbo (rather than Vceo)?

.MODEL QBC547A/PLP_XN NPN(IS=15.33f NF=1.002 ISE=0.7932f NE=1.436
BF=178.7 IKF=121.6m VAF=69.7 NR=1.004
+ ISC=83.05f NC=1.207 BR=8.628 IKR=112.1m VAR=44.7 RB=1 IRB=1u RBM=1
+ RE=639.5m RC=650.8m XTB=0 EG=1.11 XTI=3 CJE=16.1p VJE=420.9m
MJE=307.1m
+ TF=499.5p XTF=139 VTF=3.523 ITF=702.1m PTF=0 CJC=4.388p VJC=200m
MJC=279.3m
+ XCJC=619.3m TR=1.e-032 CJS=0 VJS=750m MJS=333m FC=776.2m)

.MODEL QBC557A/PLP_XN PNP(IS=20.59f NF=1.003 ISE=2.971f NE=1.316
BF=227.3 IKF=87.19m VAF=37.2 NR=1.007
+ ISC=13.39f NC=1.15 BR=7.69 IKR=76.46m VAR=11.42 RB=1 IRB=1u RBM=1
+ RE=688m RC=643.7m XTB=0 EG=1.11 XTI=3 CJE=14p VJE=591.2m MJE=357.2m
+ TF=704.6p XTF=4.217 VTF=5.367 ITF=194.7m PTF=0 CJC=11.13p VJC=100m
MJC=341.4m
+ XCJC=628.8m TR=1.e-032 CJS=0 VJS=750m MJS=333m FC=794.7m)

...Jim Thompson

So for these parts the npn B-C diode is about 4x less capacitance
(4.4pF)than the B-E diode (16.1pF), whereas for the pnp they are both
similar - 11pF vs 14pF (all at zero bias)

Thanks very much
Terry

 

[Jim Thompson]

[snip]

Hi Jim,


So usually the lowest capacitance diode is the B-C junction, ie B-E
shorted, but as usual *it depends* on the actual device used


I presume the B-C junction will break down at Vcbo (rather than Vceo)?

[snip]

Yes, Provided you tie the emitter to the base. Let it float and see
what funny business you get ;-)

...Jim Thompson

 

[Terry Given]

[snip]

Hi Jim,



So usually the lowest capacitance diode is the B-C junction, ie B-E
shorted, but as usual *it depends* on the actual device used



I presume the B-C junction will break down at Vcbo (rather than Vceo)?

[snip]

Yes, Provided you tie the emitter to the base. Let it float and see
what funny business you get ;-)

...Jim Thompson

Hi Jim,

Thanks for that.

Cheers
Terry

 

[Pooh Bear]

There is a TO-220 version of the LM35. Handy on heat sinks.

A bit pricier than the device I was using though ! They produced pretty
reproducible results ( good for about +/- 5C ).

Graham

 

[Jim Thompson]

[snip]

Hi Jim,


CJE is generally larger than CJC (see below), but not necessarily so,
since capacitance is a function of both doping level _and_ area.

So usually the lowest capacitance diode is the B-C junction, ie B-E
shorted, but as usual *it depends* on the actual device used


A B-E junction has lower breakdown than B-C.

I presume the B-C junction will break down at Vcbo (rather than Vceo)?

[snip]

Yes, Provided you tie the emitter to the base. Let it float and see
what funny business you get ;-)

...Jim Thompson

Hi Jim,

Thanks for that.

Cheers
Terry

You are quite welcome! I hope I was helpful.

...Jim Thompson

 

[John Larkin]

I've forward-biased the input ESD diode in a variety of semiconductors
(micro, CPLD etc) to measure Tj.

Cheers
Terry

You can also make quick time-slice measurements of the substrate diode
of a power mosfet to measure realtime junction temp and deduce
transient thermal behavior.

John

 

[Terry Given]

You can also make quick time-slice measurements of the substrate diode
of a power mosfet to measure realtime junction temp and deduce
transient thermal behavior.

John

Yep. And it also works with 600A IGBTs. Very important when you beat the
hell out of your silicon

I did an analysis recently of a DCM flyback converter using an infineon
FET. My simulation closely matched the real waveforms, but I found a
start-up problem (measured it too) which gave the FET a hiding. Infineon
kindly supplied me with a 6th order thermal model, which was duly
entered into the simulation, along with a function block to measure
power loss. The work of a few minutes, but it then allowed me to look at
the effects of various soft-start circuits on dTj (conclusion: the
soft-start quartered peak dTj)

Do you have any data on thermal cycling related fatigue of little
devices (T0220, TO3P etc)? Big IGBT manufacturers have these nice
lifetime curves - no. of operations versus delta-Tj (eg powerex have
them in most datasheets). The failure mechanism is thermal runaway -
large dTj plus differing CTE causes voids under the die, increasing
Rtheta, leading to more hot spots...voila. Little dutch time bomb, tick
tock boom. see http://www.pwrx.com/pwrx/app/IGBT-Intelligent-PwrMods.pdf
figure 3.16, p.15:

6,000 cycles at dTj = 100K;
200,000 cycles at dTj = 50K;
1,000,000 cycles at dTj = 30K

so the no. of cycles halves for a 10K increase in dTj.

I once designed a constant-loss modulator - it made switching frequency
inversely proportional (ish - curve fit) to load current, so at light
load there was a lot of switching loss and less conduction loss;
vice-versa at high load. That way the IGBTs didnt cool down so much
between load steps, thereby reducing dTj and dramatically increasing
lifetime.

Cheers
Terry

 

[Joerg]

Hi Dingus,

The reasons Terry gave are the most common for using a transistor as a diode. Heck, I have even used them for logic if I didn't want to spring for another CMOS chips. Whatever saves a penny is usually the better solution.

Also, if you had to demodulate or detect very tiny signals a reversed transistor exhibits a much better performance while a 'real' diode often is not very useful at the millivolt level.


Regards, Joerg

 

[Terry Given]

Hi Dingus,

The reasons Terry gave are the most common for using a transistor as a
diode. Heck, I have even used them for logic if I didn't want to spring
for another CMOS chips. Whatever saves a penny is usually the better
solution.

Yep - I've got an inverter made from a BC847bpn and a 10k 1206 quadpack
buried in the synchronous rectifier gate-drive circuitry for the 55W
smps inside each module. I also use them as gate clamping zeners (in
addition to the fast gate turn-off) in my charge-pump-driven soft-start
bypass FET. I want to try out John Larkins suggestion of E-C breakdown
(reverse-breakdown of B-E, forward bias of B-C) to see if I can throw
out a zener diode (thats biased to 5mA for low tempco) now

Also, if you had to demodulate or detect very tiny signals a reversed
transistor exhibits a much better performance while a 'real' diode often
is not very useful at the millivolt level.

Hmm, notice the terms VJE and VJC in the models Jim posted:

bc547: VJE = 420.9mV, VJC = 200mV
bc557: VJE = 591.2mV, VJC = 100mV

According to "semiconductor modelling with SPICE" (Massobrio et al)
these are the base-emitter and base-collector "built-in potentials".

They crop up in the capacitance calcs as the factor (1-Vbe/VJE)^MJE and
likewise (1-Vbc/VJC)^MJC

So the improved behaviour of the inverse-connected base-collector
junction is apparent.

Thanks Jim for the idea (doh) of looking at a spice model. Its quite
complex. Does it work for inverse connections? (methinks it should)

Question:
=========
the inverse pnp (100mV) appears better than the inverse npn (200mV),
even though the converse holds for conventional use (591mV vs 421mV). Is
this actually so?

Regards, Joerg

Cheers
Terry

 

[John Larkin]

Yep. And it also works with 600A IGBTs. Very important when you beat the
hell out of your silicon

I did an analysis recently of a DCM flyback converter using an infineon
FET. My simulation closely matched the real waveforms, but I found a
start-up problem (measured it too) which gave the FET a hiding. Infineon
kindly supplied me with a 6th order thermal model, which was duly
entered into the simulation, along with a function block to measure
power loss. The work of a few minutes, but it then allowed me to look at
the effects of various soft-start circuits on dTj (conclusion: the
soft-start quartered peak dTj)

Do you have any data on thermal cycling related fatigue of little
devices (T0220, TO3P etc)? Big IGBT manufacturers have these nice
lifetime curves - no. of operations versus delta-Tj (eg powerex have
them in most datasheets). The failure mechanism is thermal runaway -
large dTj plus differing CTE causes voids under the die, increasing
Rtheta, leading to more hot spots...voila. Little dutch time bomb, tick
tock boom. see http://www.pwrx.com/pwrx/app/IGBT-Intelligent-PwrMods.pdf
figure 3.16, p.15:

6,000 cycles at dTj = 100K;
200,000 cycles at dTj = 50K;
1,000,000 cycles at dTj = 30K

6000 cycles! Yikes!

so the no. of cycles halves for a 10K increase in dTj.

I once designed a constant-loss modulator - it made switching frequency
inversely proportional (ish - curve fit) to load current, so at light
load there was a lot of switching loss and less conduction loss;
vice-versa at high load. That way the IGBTs didnt cool down so much
between load steps, thereby reducing dTj and dramatically increasing
lifetime.

No, no data on thermal fatigue. I thought that problem was solved ages
ago. My NMR pulsed-gradient drivers have huge dissipation spikes,
typically going from, say, 40C to 130C Tj in 20-40 milliseconds, once
a second, in big 300 watt mosfets. I've worried about fatigue, but so
far haven't had many fet failures, or signs of failure rate increasing
with use. I run software that simulates Tj in real time and shuts down
the amps at simulated 140C junction temp. Why don't big fets have
on-chip temp sensors like CPUs?

Maybe bigger slabs of silicon, like huge SCRs and IGBTs, have more
severe thermal stress issues. IGBTs have real high power densities,
don't they?

I did discover that a lot of 300-watt rated power fets will explode in
under 100 msec when they really dissipate 300 watts, bolted directly
to a cold copper block.

John

 

[Terry Given]

6000 cycles! Yikes!

Its pretty bad all right. Normally in a drive this only happens when
there is an overload. But overload ratings are a great area for
specmanship. nd you can see how its quite possible to get great overload
ratings for a while...

No, no data on thermal fatigue. I thought that problem was solved ages
ago. My NMR pulsed-gradient drivers have huge dissipation spikes,
typically going from, say, 40C to 130C Tj in 20-40 milliseconds, once
a second, in big 300 watt mosfets. I've worried about fatigue, but so
far haven't had many fet failures, or signs of failure rate increasing
with use. I run software that simulates Tj in real time and shuts down
the amps at simulated 140C junction temp. Why don't big fets have
on-chip temp sensors like CPUs?

A lot of big IGBTs do. Semikron make a range of neat goodies. You might
want to take a look at their SemiTOP range.

Maybe bigger slabs of silicon, like huge SCRs and IGBTs, have more
severe thermal stress issues. IGBTs have real high power densities,
don't they?

These powerex die are about 1cm on a side. Conduction loss is 1200W or
so at 600A. Switching loss is a lot higher, but obviously for short
durations. And yeah, its certainly related to the size - dx/x is
constant but larger x has a correspondingly larger dx.

I did discover that a lot of 300-watt rated power fets will explode in
under 100 msec when they really dissipate 300 watts, bolted directly
to a cold copper block.

John

I've seen those sorts of silly numbers for TO220 parts with Rtheta_j_s
of 1K/W. Go figure.....mind you I once worked with a guy who had built a
25kW liquid-nitrogen cooled inverter using little FETs.

Semikron once sold us a "rupture proof IGBT package" so we smacked it
into a prototype 100kW inverter DC bus assembly and dumped a full bus
charge into it - 1.4kJ or so. It went BANG and the whole assembly jumped
a few feet in the air. So we replaced the dead-but-not-ruptured IGBT
with another one, and then sat an anvil on top of the PCB. This time the
package exploded - in front of the salesman. So much for rupture-proof

Mind you I once tested an smt 1500W TVS with a 600J pulse. There was
NOTHING left - not even the J-leads. I had covered the experiment with a
pyrex jug, and there were several distinctive splats - copper,
carbonised plastic and re-solidified silicon. The office lady was pissed
I had wrecked the jug, so I bought her a new one

Cheers
Terry