Fast switching current mirrors

[David Moreno]

Hi !

I'm currently working in a fast switching current mirror to charge
linearly a capacitor. The circuit is a normal current mirror with
a multiplexor attached to the base of the second transistor and a
voltage
generator to switch the current throught it base.

In the simulation I've found that when you rise the frequency (at
50MHz or so)
the current mirror is not fast enought to switch, and a delay appears
that
makes the current throught the capacitor to lower. I'm intriguing in
this
because the transistor that I'm using are known to work up to 5-6GHz,
and the
frecuency of the switching is well below that.

In the practice the results are simply that the current mirrored is
not
the same as the origin current, and the diference increases when you
rise
the frecuency.

And finally the question: Some way to make this work faster or better
the response at high frecuencies ?

Thanks a lot ! =D

--------------------------------------------------------------------------------------
* E:\ALIENFILES\electronica\current mirror\espejo bft93.asc
X§Q1 N003 N003 N001 BFT93
X§Q2 Uc N004 N001 BFT93
C1 Uc 0 47pF
V1 N001 0 5
I1 N003 0 PULSE(0 1mA 0 0.1ns 0.1ns 5ns 10ns)
V2 N002 0 PULSE(0 5 0 0.1ns 0.1ns 5ns 10ns)
S1 N003 N004 N002 0 miswitch
S2 N004 N001 0 N002 miswitch
..inc "E:\ALIENFILES\electronica\spice\bft93.spice"
..tran 240n
..model miswitch SW(Ron=1. Roff=1MEG Vt=2.5V Vh=0 Lser=0 Vser=0)
..ic V(Uc)=0V
..backanno
..end
--------------------------------------------------------------------------------------------

bft93.spice

* Filename: BFT93_SPICE.PRM
* BFT93 SPICE MODEL
* PHILIPS SEMICONDUCTORS
* Date : September 1995
*
* PACKAGE : SOT23 DIE MODEL : BFT93
* 1: COLLECTOR; 2: BASE; 3: EMITTER;
..SUBCKT BFT93 1 2 3
Q1 6 5 7 7 BFT93
* SOT23 parasitic model
Lb 4 5 .4nH
Le 7 8 .83nH
L1 2 4 .35nH
L2 1 6 .17nH
L3 3 8 .35nH
Ccb 4 6 0.071pF
Cbe 4 8 0.002pF
Cce 6 8 0.071pF
* PHILIPS SEMICONDUCTORS Version:
2.0
* Filename: BFT93.PRM Date:
March 1992
*
..MODEL BFT93 PNP
+ IS = 8.35127E-016
+ BF = 4.85648E+001
+ NF = 1.00043E+000
+ VAF = 1.90118E+001
+ IKF = 1.46824E-001
+ ISE = 9.09455E-014
+ NE = 1.74928E+000
+ BR = 1.21832E+001
+ NR = 9.97694E-001
+ VAR = 3.37492E+000
+ IKR = 6.74270E-003
+ ISC = 2.34297E-014
+ NC = 1.44993E+000
+ RB = 1.00000E+001
+ IRB = 1.00000E-006
+ RBM = 1.00000E+001
+ RE = 2.00000E-001
+ RC = 3.80000E+000
+ EG = 1.11000E+000
+ XTI = 3.00000E+000
+ CJE = 1.57034E-012
+ VJE = 6.00000E-001
+ MJE = 3.82204E-001
+ TF = 1.48531E-011
+ XTF = 2.20970E+000
+ VTF = 2.98987E+000
+ ITF = 1.43721E-002
+ CJC = 1.99525E-012
+ VJC = 5.84499E-001
+ MJC = 2.81320E-001
..ENDS
------------------------------------------------------------

 

[Kevin Aylward]

David Moreno wrote:

Hi !

I'm currently working in a fast switching current mirror to charge
linearly a capacitor. The circuit is a normal current mirror with
a multiplexor attached to the base of the second transistor and a
voltage
generator to switch the current throught it base.

In the simulation I've found that when you rise the frequency (at
50MHz or so)
the current mirror is not fast enought to switch, and a delay appears
that
makes the current throught the capacitor to lower. I'm intriguing in
this
because the transistor that I'm using are known to work up to 5-6GHz,
and the
frecuency of the switching is well below that.

The ft is the frequency that the current gain is unity. The switching
speed of a transiser can be orders below this. Ft is only a guide.

In the practice the results are simply that the current mirrored is
not
the same as the origin current, and the diference increases when you
rise
the frecuency.

And finally the question: Some way to make this work faster or better
the response at high frecuencies ?

You havent posted the actuall circuit (LTSpice?), but from the text it
looks like your switching the base. I am not going to wade through a
text schematic foir a better evaluation.

The fastest way to switch (usually) is to current steer the *output*
current, not the base current. That is, keep the mirror always on and
divert the output from the required output to a dummy output. A mos is a
reasonable way to do this.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[David Moreno]

Hi!

and the
frecuency of the switching is well below that.

The ft is the frequency that the current gain is unity. The switching
speed of a transiser can be orders below this. Ft is only a guide.

Ok, Which parameter do you think is most important then for a switching
current ?

You havent posted the actuall circuit (LTSpice?), but from the text it

Yes, LTSPICE. By post do you mean in binaries, no ?

looks like your switching the base. I am not going to wade through a
text schematic foir a better evaluation.

Yep, sorry U_U. Next time I'll post the file.

The fastest way to switch (usually) is to current steer the *output*
current, not the base current. That is, keep the mirror always on and
divert the output from the required output to a dummy output. A mos is a
reasonable way to do this.

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is not
really recomendable.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

I've to try this Spice also =D

 

[John Larkin]

On 25 Jun 2004 14:47:08 -0700, saimazoom@hotmail.com (David Moreno)
wrote:

Hi!

and the
frecuency of the switching is well below that.

The ft is the frequency that the current gain is unity. The switching
speed of a transiser can be orders below this. Ft is only a guide.

Ok, Which parameter do you think is most important then for a switching
current ?

You havent posted the actuall circuit (LTSpice?), but from the text it

Yes, LTSPICE. By post do you mean in binaries, no ?

looks like your switching the base. I am not going to wade through a
text schematic foir a better evaluation.

Yep, sorry U_U. Next time I'll post the file.

The fastest way to switch (usually) is to current steer the *output*
current, not the base current. That is, keep the mirror always on and
divert the output from the required output to a dummy output. A mos is a
reasonable way to do this.

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is not
really recomendable.

I've done this a bunch of times. Start with a precise, slow current
source, maybe a high-beta PNP transistor inside an opamp feedback
loop. That dumps current into a PNP differential pair which steers the
current into your capacitor or dumps it to ground; wiggle the diff
pair bases to steer; I'm usually using differential ECL here. For even
more fun, use the "dumped" current to turn on a transistor that
discharges the cap when the current source is switched "off". Charge
injection is minimal.

You need high-beta transistors to avoid base current:temperature
errors. Microwave PNPs are tempting, but tend to have low betas and
like to oscillate in this setup. BCX71K is a nice part, medium-fast
and high beta. I've done a nice 13 ns linear ramp this way, with a few
additional tricks.

What are you building? How fast?

John

 

[Kevin Aylward]

David Moreno wrote:

Hi!

and the
frecuency of the switching is well below that.

The ft is the frequency that the current gain is unity. The switching
speed of a transiser can be orders below this. Ft is only a guide.

Ok, Which parameter do you think is most important then for a
switching current ?

The Ft is good guide. There is no one parameter. Everything is a trade
off.

You havent posted the actuall circuit (LTSpice?), but from the text
it

Yes, LTSPICE. By post do you mean in binaries, no ?

No, I meant the LTSpice full text schematic file.

looks like your switching the base. I am not going to wade through a
text schematic foir a better evaluation.

Yep, sorry U_U. Next time I'll post the file.

The fastest way to switch (usually) is to current steer the *output*
current, not the base current. That is, keep the mirror always on and
divert the output from the required output to a dummy output. A mos
is a reasonable way to do this.

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially.

Yes.

Serial switching is not
really recommendable.

Of course it is. It is a very useful technique and used a lot, e.g.
current steered DACs. Whether or not there is significant charge
injection depends on the particular application. I have certainly done
this in a charge pump for a PLL where this method was the only one that
achieved the required speed and accuracy.

*Whenever* you do a switch, there will be always be some level of charge
injection. For example, if you switch the base there will be feedthrough
via Cbc.

Best Regards,

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[David Moreno]

Hi !

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is not
really recomendable.


I've done this a bunch of times. Start with a precise, slow current
source, maybe a high-beta PNP transistor inside an opamp feedback
loop. That dumps current into a PNP differential pair which steers the
current into your capacitor or dumps it to ground; wiggle the diff
pair bases to steer; I'm usually using differential ECL here. For even
more fun, use the "dumped" current to turn on a transistor that
discharges the cap when the current source is switched "off". Charge
injection is minimal.

That's exactly the next circuit that I'm goind to work into. I'll put
two series resistor to excite the base of the second transistor, and
introduce
some square voltage signal (arount 1Vpp) in the base of first
transistor.

I've simulated this topology and It seems to work well, but when you
raise the frecuency the circuit seems to no respond well (too slow).

You need high-beta transistors to avoid base current:temperature
errors. Microwave PNPs are tempting, but tend to have low betas and
like to oscillate in this setup. BCX71K is a nice part, medium-fast
and high beta. I've done a nice 13 ns linear ramp this way, with a few
additional tricks.

That transistor is too slow for my purposes (around 100MHZ) but I'll
give
it a Try with BFT93. Thanks for the Help ! =D

What are you building? How fast?

I'm currently working in Hardware Neural Networks. I'm using this
current
mirror to store contributions in a capacitor (as a local memory). I've
a working circuit at 1MHz or so, but now I'm trying to increase the
working frecuency.

Un saludo.

 

[David Moreno]

"Kevin Aylward" <kevin.aylwardEXTRACT@anasoft.co.uk> wrote in message news:<IM8Dc.3475$T3.2096@front-1.news.blueyonder.co.uk>...

The Ft is good guide. There is no one parameter. Everything is a trade
off.

There is no shortcut then to select the adecuate transistor. If I
guide
by the Ft parameter then my transistor are ok for my purposes.

You havent posted the actuall circuit (LTSpice?), but from the text
it

Yes, LTSPICE. By post do you mean in binaries, no ?

No, I meant the LTSpice full text schematic file.

Ah, It was directly in the message. I've posted also the transistor
SPICE model, look down in the text of the original post.

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially.

Yes.

That's something I try to reduce.

Serial switching is not
really recommendable.

Of course it is. It is a very useful technique and used a lot, e.g.
current steered DACs. Whether or not there is significant charge

Sorry, I'll try to fix it: Is not really recomendable if you really
mind about the current injection.

injection depends on the particular application. I have certainly done

You are absolutely right.

this in a charge pump for a PLL where this method was the only one that
achieved the required speed and accuracy.

At which frecuency ?. I had a very hard time finding analog switches
as fast as 100MHz, finally found them in Texas Instruments. And the
injection
charge is not negligble at all if you put them serrially with your
load (a capacitor in my case).

*Whenever* you do a switch, there will be always be some level of charge
injection. For example, if you switch the base there will be feedthrough
via Cbc.

But is something that, in the practical circuit, isn't as bad as it
sounds. Finally I raised the current, and the circuit worked pretty
well but what if I want to use lower currents ?.

Best Regards,

Thanks for your help ! =)

 

[John Larkin]

On 27 Jun 2004 13:31:55 -0700, saimazoom@hotmail.com (David Moreno)
wrote:

Hi !

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is not
really recomendable.


I've done this a bunch of times. Start with a precise, slow current
source, maybe a high-beta PNP transistor inside an opamp feedback
loop. That dumps current into a PNP differential pair which steers the
current into your capacitor or dumps it to ground; wiggle the diff
pair bases to steer; I'm usually using differential ECL here. For even
more fun, use the "dumped" current to turn on a transistor that
discharges the cap when the current source is switched "off". Charge
injection is minimal.

That's exactly the next circuit that I'm goind to work into. I'll put
two series resistor to excite the base of the second transistor, and
introduce
some square voltage signal (arount 1Vpp) in the base of first
transistor.

I've simulated this topology and It seems to work well, but when you
raise the frecuency the circuit seems to no respond well (too slow).

You need high-beta transistors to avoid base current:temperature
errors. Microwave PNPs are tempting, but tend to have low betas and
like to oscillate in this setup. BCX71K is a nice part, medium-fast
and high beta. I've done a nice 13 ns linear ramp this way, with a few
additional tricks.

That transistor is too slow for my purposes (around 100MHZ) but I'll
give
it a Try with BFT93. Thanks for the Help ! =D

What are you building? How fast?

I'm currently working in Hardware Neural Networks. I'm using this
current
mirror to store contributions in a capacitor (as a local memory). I've
a working circuit at 1MHz or so, but now I'm trying to increase the
working frecuency.

Un saludo.

Has there ever been a neural network that actually did something
useful? I know a few nn enthusiasts, scientists and not engineers, who
keep suggesting it be used to solve signal processing problems where
I'm certain it could never work. And so far, it hasn't

Seems awfully unscientific to me.

John

 

[John Larkin]

On 27 Jun 2004 13:31:55 -0700, saimazoom@hotmail.com (David Moreno)
wrote:

Hi !

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is not
really recomendable.


I've done this a bunch of times. Start with a precise, slow current
source, maybe a high-beta PNP transistor inside an opamp feedback
loop. That dumps current into a PNP differential pair which steers the
current into your capacitor or dumps it to ground; wiggle the diff
pair bases to steer; I'm usually using differential ECL here. For even
more fun, use the "dumped" current to turn on a transistor that
discharges the cap when the current source is switched "off". Charge
injection is minimal.

That's exactly the next circuit that I'm goind to work into. I'll put
two series resistor to excite the base of the second transistor, and
introduce
some square voltage signal (arount 1Vpp) in the base of first
transistor.

I've simulated this topology and It seems to work well, but when you
raise the frecuency the circuit seems to no respond well (too slow).

You need high-beta transistors to avoid base current:temperature
errors. Microwave PNPs are tempting, but tend to have low betas and
like to oscillate in this setup. BCX71K is a nice part, medium-fast
and high beta. I've done a nice 13 ns linear ramp this way, with a few
additional tricks.

That transistor is too slow for my purposes (around 100MHZ) but I'll
give
it a Try with BFT93. Thanks for the Help ! =D

What are you building? How fast?

I'm currently working in Hardware Neural Networks. I'm using this
current
mirror to store contributions in a capacitor (as a local memory). I've
a working circuit at 1MHz or so, but now I'm trying to increase the
working frecuency.

Un saludo.

Oh, this box

http://www.signalrecovery.com/9650a.htm

uses differential PNP current steering to create interruptable linear
ramps; it's based on Pepper's fiendishly clever (but fatally flawed)
delay vernier patent. I think the clock period here is 80 or maybe 120
MHz, and they start/stop ramps on clock edges, with picosecond
precision.

John

 

[Kevin Aylward]

David Moreno wrote:

"Kevin Aylward" <kevin.aylwardEXTRACT@anasoft.co.uk> wrote in message
news:<IM8Dc.3475$T3.2096@front-1.news.blueyonder.co.uk>...


depends on the particular application. I have certainly
done

You are absolutely right.

this in a charge pump for a PLL where this method was the only one
that achieved the required speed and accuracy.

At which frecuency ?.

Nominal 100Mhz, but the switching speed had to be << 1ns.

I had a very hard time finding analog switches
as fast as 100MHz,

This was a 0.18u i.c. cmos process

finally found them in Texas Instruments. And the
injection
charge is not negligble at all if you put them serrially with your
load (a capacitor in my case).

*Whenever* you do a switch, there will be always be some level of
charge injection. For example, if you switch the base there will be
feedthrough via Cbc.

But is something that, in the practical circuit, isn't as bad as it
sounds. Finally I raised the current, and the circuit worked pretty
well but what if I want to use lower currents ?.

Its the trade off thing again g(power, sigma, frequency) = 0

http://www.anasoft.co.uk/EE/bipolarpafl/bipolarpafl.html

Lower power means something *has* to give, either speed or accuracy.
There is no such thing as a free lunch.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[John Larkin]

On Tue, 29 Jun 2004 20:14:34 +0100, "Kevin Aylward"
<salesEXTRACT@anasoft.co.uk> wrote:

Its the trade off thing again g(power, sigma, frequency) = 0

http://www.anasoft.co.uk/EE/bipolarpafl/bipolarpafl.html

Lower power means something *has* to give, either speed or accuracy.
There is no such thing as a free lunch.

Then there's the very zen-like distributed amplifier that puts
transistors in parallel without putting them in parallel. This
transcends Gm/c limits; doesn't work with bipolars, though.

John

 

[Kevin Aylward]

John Larkin wrote:

On Tue, 29 Jun 2004 20:14:34 +0100, "Kevin Aylward"
salesEXTRACT@anasoft.co.uk> wrote:




Its the trade off thing again g(power, sigma, frequency) = 0

http://www.anasoft.co.uk/EE/bipolarpafl/bipolarpafl.html

Lower power means something *has* to give, either speed or accuracy.
There is no such thing as a free lunch.


Then there's the very zen-like distributed amplifier that puts
transistors in parallel without putting them in parallel. This
transcends Gm/c limits; doesn't work with bipolars, though.

I am only vaguely familiar with this method, but I very much doubt it
gets around what I posted. Its not a direct Gm/c limit. Its a combined
power, speed, accuracy limit. The "fundamental" physics limit, in
principle allows any speed, providing one sacrifices power or accuracy,
or both. adding devices will increase the current.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[David Moreno]

John Larkin <jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> wrote in message news:<68mud0po80dccqukulrkkj862hat8vso9i@4ax.com>...

Has there ever been a neural network that actually did something
useful? I know a few nn enthusiasts, scientists and not engineers, who

There are a few, ie: www.neuricam.com

keep suggesting it be used to solve signal processing problems where
I'm certain it could never work. And so far, it hasn't

Well, I've already build a prototype that is able of recognize
images with a high level of noise in them. Now I'm more concerned
to build a basic fast node, then I'll se how to apply it to image
processing.

Seems awfully unscientific to me.

Why do you say that ?. Neural networks has been a scientific threat
for about 60 years =).


Un saludo

 

[David Moreno]

"Kevin Aylward" <salesEXTRACT@anasoft.co.uk> wrote in message news:<tcjEc.2837$a37.2677@fe2.news.blueyonder.co.uk>...

At which frecuency ?.
Nominal 100Mhz, but the switching speed had to be << 1ns.
I had a very hard time finding analog switches
as fast as 100MHz,

This was a 0.18u i.c. cmos process

Wow, the skies of electronic design ... =). Right now I'm
using FiberGlass and Taconic Boards.

Its the trade off thing again g(power, sigma, frequency) = 0

http://www.anasoft.co.uk/EE/bipolarpafl/bipolarpafl.html

I'll take a look at this, I hope this will answer me some questions. Thanks !

Lower power means something *has* to give, either speed or accuracy.
There is no such thing as a free lunch.

I want free lunch ! X)

Un saludo

 

[Kevin Aylward]

John Larkin wrote:

On 27 Jun 2004 13:31:55 -0700, saimazoom@hotmail.com (David Moreno)
wrote:

Hi !

Yes, but when you do that the switching element injects a charge
in the capacitor that discharge it parcially. Serial switching is
not really recomendable.


I've done this a bunch of times. Start with a precise, slow current
source, maybe a high-beta PNP transistor inside an opamp feedback
loop. That dumps current into a PNP differential pair which steers
the current into your capacitor or dumps it to ground; wiggle the
diff pair bases to steer; I'm usually using differential ECL here.
For even more fun, use the "dumped" current to turn on a transistor
that discharges the cap when the current source is switched "off".
Charge injection is minimal.

That's exactly the next circuit that I'm goind to work into. I'll put
two series resistor to excite the base of the second transistor, and
introduce
some square voltage signal (arount 1Vpp) in the base of first
transistor.

I've simulated this topology and It seems to work well, but when you
raise the frecuency the circuit seems to no respond well (too slow).

You need high-beta transistors to avoid base current:temperature
errors. Microwave PNPs are tempting, but tend to have low betas and
like to oscillate in this setup. BCX71K is a nice part, medium-fast
and high beta. I've done a nice 13 ns linear ramp this way, with a
few additional tricks.

That transistor is too slow for my purposes (around 100MHZ) but I'll
give
it a Try with BFT93. Thanks for the Help ! =D

What are you building? How fast?

I'm currently working in Hardware Neural Networks. I'm using this
current
mirror to store contributions in a capacitor (as a local memory).
I've a working circuit at 1MHz or so, but now I'm trying to increase
the working frecuency.

Un saludo.

Has there ever been a neural network that actually did something
useful?

Yes. John Larkin is one such example.

I know a few nn enthusiasts, scientists and not engineers, who

keep suggesting it be used to solve signal processing problems where

You mean you always fail to notice when the blond bit gives you the come
on message?

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[John Larkin]

On Thu, 01 Jul 2004 18:57:45 GMT, "Kevin Aylward"
<salesEXTRACT@anasoft.co.uk> wrote:

Has there ever been a neural network that actually did something
useful?

Yes. John Larkin is one such example.

Are you implying that I might be useful? I thought I was purely
decorative.

John