5mW boost converter

[Adam S]

Hi,
I am trying to run a LCD character module from a 3.3V supply in a
battery powered instrument. The LCD module logic runs perfectly at 3.3V
(tested it down to 1.8V), however the LCD display voltage needs about
4.5~5V negative to VDD (@ 1mA). Normally a simple charge pump would do,
but I need to save every milliamp possible. This is where I thought up a
simple boost converted like shown in the following schematic.

http://members.optusnet.com.au/~eseychell/pictures/boost_conv_01.png

Measured efficiency of this converter was about 60% with a 1mA load
between +3.3V and VOUT as shown. I tried to find where most of the power
is lost, and it seems due to the 74HC14. With 5V applied across the
74HC14, the current into the VDD pin is 0.8mA, and when I remove L1 it
drops further to 0.6mA. Any ideas on reducing oscillator current while
still keeping it fairly simple ?

Adam

 

[Bill Sloman]

Hi,
I am trying to run a LCD character module from a 3.3V supply in a battery
powered instrument. The LCD module logic runs perfectly at 3.3V (tested it
down to 1.8V), however the LCD display voltage needs about 4.5~5V negative
to VDD (@ 1mA). Normally a simple charge pump would do, but I need to save
every milliamp possible. This is where I thought up a simple boost
converted like shown in the following schematic.

http://members.optusnet.com.au/~eseychell/pictures/boost_conv_01.png

Measured efficiency of this converter was about 60% with a 1mA load
between +3.3V and VOUT as shown. I tried to find where most of the power
is lost, and it seems due to the 74HC14. With 5V applied across the
74HC14, the current into the VDD pin is 0.8mA, and when I remove L1 it
drops further to 0.6mA. Any ideas on reducing oscillator current while
still keeping it fairly simple ?

The big weakness of CMOS in this sort of application is that both the P and
N-channel MOSFETs are on during switchover, so you get some "shoot-through"
current. If you used discrete logic-level P- and N-channel MOSFETs to drive
the inductor, and organised the logic to give you non-over-lapping drive
waveforms, you might do better - the circuit might not have to be much more
complicated.

Or you can replace your inductor with a transformer wound on an RM4 cores,
or one of the little planar cores, and use something really simple. like a
Royer inverter, which gets up to around 90% efficiency in some situations.

 

[John Popelish]

... Any ideas on reducing oscillator current while
still keeping it fairly simple ?

Are you willing to consider other approaches?
http://cache.national.com/ds/LM/LM2664.pdf

 

[Adam S]

The big weakness of CMOS in this sort of application is that both the P and
N-channel MOSFETs are on during switchover, so you get some "shoot-through"
current. If you used discrete logic-level P- and N-channel MOSFETs to drive
the inductor, and organised the logic to give you non-over-lapping drive
waveforms, you might do better - the circuit might not have to be much more
complicated.

Or you can replace your inductor with a transformer wound on an RM4 cores,
or one of the little planar cores, and use something really simple. like a
Royer inverter, which gets up to around 90% efficiency in some situations.

I was thinking about the shoot through issue, and wondered whether
changing to 40106 CMOS hex Schmitt trigger will make a difference. Then
I discovered that the bias current for the LCD character module was
essentially due to an onboard voltage divider comprised of 5 x 1 kohm
resistors. Changing these to 10kohm will 1/10th the current consumption.
At 100uA, a simple capacitor switcher such as the LM6642 that John
Popelish suggested looks attractive.

Adam

 

[Fritz Schlunder]

Hi,
I am trying to run a LCD character module from a 3.3V supply in a
battery powered instrument. The LCD module logic runs perfectly at 3.3V
(tested it down to 1.8V), however the LCD display voltage needs about
4.5~5V negative to VDD (@ 1mA). Normally a simple charge pump would do,
but I need to save every milliamp possible. This is where I thought up a
simple boost converted like shown in the following schematic.

http://members.optusnet.com.au/~eseychell/pictures/boost_conv_01.png

Measured efficiency of this converter was about 60% with a 1mA load
between +3.3V and VOUT as shown. I tried to find where most of the power
is lost, and it seems due to the 74HC14. With 5V applied across the
74HC14, the current into the VDD pin is 0.8mA, and when I remove L1 it
drops further to 0.6mA. Any ideas on reducing oscillator current while
still keeping it fairly simple ?

Adam

The CMOS 555 may perform somewhat better in this application. Most of your
current power waste is probably being lost due to cross conduction of the
input buffer stage of the U1A inverter. Unfortunately this type of
oscillator requires the input voltage to be roughly 1/2 the chip voltage,
which maximizes cross conduction waste.

 

[[email protected]]

Hi,
I am trying to run a LCD character module from a 3.3V supply in a
battery powered instrument. The LCD module logic runs perfectly at 3.3V
(tested it down to 1.8V), however the LCD display voltage needs about
4.5~5V negative to VDD (@ 1mA). Normally a simple charge pump would do,
but I need to save every milliamp possible. This is where I thought up a
simple boost converted like shown in the following schematic.

http://members.optusnet.com.au/~eseychell/pictures/boost_conv_01.png

Measured efficiency of this converter was about 60% with a 1mA load
between +3.3V and VOUT as shown. I tried to find where most of the power
is lost, and it seems due to the 74HC14. With 5V applied across the
74HC14, the current into the VDD pin is 0.8mA, and when I remove L1 it
drops further to 0.6mA. Any ideas on reducing oscillator current while
still keeping it fairly simple ?

Adam

I'd wager the problem isn't shoot-through on the paralleled stages,
but cross-conduction in the oscillator stage. I've measured
disappointingly high Idd in 74hc14s I'd hoped to use as micro-power
oscillators. They aren't suitable (Idd close to a mA for slow
oscillators ISTR).

The solution is to use something other than the 74hc14 for the
oscillator, something with sharp edges to feed into the CMOS part.

A much lower frequency, low-duty cycle pulse for a driving waveform
would save a bunch on switching losses too (but kills your synchronous
rectifier scheme, and you'd need feedback for voltage regulation).

Outputting 1mA @ 5v and 60% efficiency should require 2.5mA @ 3.3v
input into the 74hc14, so for the Idd(74hc14)=0.8mA you report the
converter isn't loaded, yes? Still, 0.6mA of losses @ 3.3v with L1
disconnected leaves 0.4mA of losses unaccounted for. That might be the
paralleled driver stages -- you could disable them, test, and thus be
sure.

James Arthur

 

[[email protected]]

I was thinking about the shoot through issue, and wondered whether
changing to 40106 CMOS hex Schmitt trigger will make a difference. Then
I discovered that the bias current for the LCD character module was
essentially due to an onboard voltage divider comprised of 5 x 1 kohm
resistors. Changing these to 10kohm will 1/10th the current consumption.
At 100uA, a simple capacitor switcher such as the LM6642 that John
Popelish suggested looks attractive.

Adam

Yes, just saw this. If you can live with its -3v output (or can
afford to cascade two), a capacitive switcher sounds simple, and is
spunky enough for this load.

ISTR some regulating gadgets rather like this one, but which have
multiple stages and can output double the input voltage. Not
super-efficient, but simple and solid. Made by LTC perhaps? Yes:
http://www.linear.com/pc/categoryProducts.do

James Arthur

 

[Adam S]

I'd wager the problem isn't shoot-through on the paralleled stages,
but cross-conduction in the oscillator stage. I've measured
disappointingly high Idd in 74hc14s I'd hoped to use as micro-power
oscillators. They aren't suitable (Idd close to a mA for slow
oscillators ISTR).

Thats agrees precisely with my findings too. For the 74HC14 oscillator
the VDD pin current drain is essentially independent of frequency until
you get >> 500kHz. My conclusion is cross conduction on the FET input
stage of the oscillator inverter. So your right, the 74HC14 cannot be
use as a micro power oscillator. My previous attempts were with three
gate RC oscillators using a 74HC04, but IDD was around 1.0mA. Even worse
was 3~4mA from the 74AC04. I was scratching my head for a while until I
realized I should be using a single gate Schmitt 74HC14 oscillator, but
that turned out less than ideal.

The solution is to use something other than the 74hc14 for the
oscillator, something with sharp edges to feed into the CMOS part.

I measured 200uA supply current of a 555 CMOS configures as 50% duty
cycle oscillator running at 500kHz. To get self starting synchronous
rectification, paralleled 74HC04 hex inverters will need to be added on
the output.

 

[[email protected]]

Thats agrees precisely with my findings too. For the 74HC14 oscillator
the VDD pin current drain is essentially independent of frequency until
you get >> 500kHz. My conclusion is cross conduction on the FET input
stage of the oscillator inverter. So your right, the 74HC14 cannot be
use as a micro power oscillator. My previous attempts were with three
gate RC oscillators using a 74HC04, but IDD was around 1.0mA. Even worse
was 3~4mA from the 74AC04. I was scratching my head for a while until I
realized I should be using a single gate Schmitt 74HC14 oscillator, but
that turned out less than ideal.


I measured 200uA supply current of a 555 CMOS configures as 50% duty
cycle oscillator running at 500kHz.

To get self starting synchronous
rectification, paralleled 74HC04 hex inverters will need to be added on
the output.

And maybe some provision to make sure -VOUT doesn't charge up to
+3.3v on startup, perhaps a schottky to GND?

As seductive as synchronous rectification is, in this instance the
inductor has ~8 volts flyback across it, so a schottky rectifier's 0.4v
drop will only contribute ~ 0.4v/8v = 5% overall efficiency loss. It's
pretty easy to lose that and more in switching losses in the sync.
rectifier, which already has an ohmic loss as well.

But, all these are moot: you've already got the thing beat just by
upping the LCD's divider resistances 10x. Touché.

James Arthur

 

[qrk]

Thats agrees precisely with my findings too. For the 74HC14 oscillator
the VDD pin current drain is essentially independent of frequency until
you get >> 500kHz. My conclusion is cross conduction on the FET input
stage of the oscillator inverter. So your right, the 74HC14 cannot be
use as a micro power oscillator. My previous attempts were with three
gate RC oscillators using a 74HC04, but IDD was around 1.0mA. Even worse
was 3~4mA from the 74AC04. I was scratching my head for a while until I
realized I should be using a single gate Schmitt 74HC14 oscillator, but
that turned out less than ideal.


I measured 200uA supply current of a 555 CMOS configures as 50% duty
cycle oscillator running at 500kHz. To get self starting synchronous
rectification, paralleled 74HC04 hex inverters will need to be added on
the output.

I recently did a driver circuit for LCD shutters using a LTC6906 for
an oscillator and a 4060 to provide drive for a capacitor voltage
doubler (very efficient), LCD shutter drive signal, and other timing.
A couple other parts are used to do level translation and drive the
LCD shutters. The LTC6906 draws around 15 uA at an output freq of 13
kHz. LTC6906 is one of Linear Technology's wonderful resistor-set
oscillators. Total current consumption of the circuit drew; while
driving a 10nF shutter capacitance; 25 uA in-phase, 320 uA
out-of-phase.

Making R-C oscillators out of various standard logic family inverters
consumes 100s of uA at 10kHz, partly due to the slow rise time on the
input.

 

[legg]

snip

but I need to save every milliamp possible. This is where I thought up a
simple boost converted like shown in the following schematic.

http://members.optusnet.com.au/~eseychell/pictures/boost_conv_01.png

Measured efficiency of this converter was about 60% with a 1mA load
between +3.3V and VOUT as shown. I tried to find where most of the power
is lost, and it seems due to the 74HC14. With 5V applied across the
74HC14, the current into the VDD pin is 0.8mA, and when I remove L1 it
drops further to 0.6mA. Any ideas on reducing oscillator current while
still keeping it fairly simple ?

At this low power drain, most of the energy in the circuit should be
just circulating. Peak to peak magnetizing current in the choke will
be in the order of 2mA ppk, with the values shown, but this should
mostly be returned to the rails. The 100ohm output driver resistance
is the source of loss in the actual energy conversion.

If you dropped the oscillator frequency, the parasitic
cross-conduction in the fets of the paralleled output structure should
reduce dramatically. This will be much more than is present in
independantly conneced gates.

Freewheeling magnetizing current might be more efficiently returned to
the rails using shottky signal diodes, if the inductor value isn't
raised accordingly, to keep magnetization constant.

By the way, I think you'll find it works just as well without biasing
the duty cycle away from 50% ( ie without R2).

RL

 

[Joerg]

Hello Adam,

Thats agrees precisely with my findings too. For the 74HC14 oscillator
the VDD pin current drain is essentially independent of frequency until
you get >> 500kHz. My conclusion is cross conduction on the FET input
stage of the oscillator inverter. So your right, the 74HC14 cannot be
use as a micro power oscillator. My previous attempts were with three
gate RC oscillators using a 74HC04, but IDD was around 1.0mA. Even worse
was 3~4mA from the 74AC04. I was scratching my head for a while until I
realized I should be using a single gate Schmitt 74HC14 oscillator, but
that turned out less than ideal.

Strange, I have used the CD40106 as well as the 74HC14 in low power PWM
converters without any problems. They are both Schmitts, meaning they
won't dwell in the crossover phase any longer than their switching speed
lets them. They only consumed more when I went above a MHz or so but
that is to be expected from older CMOS processes.

But they do not like stuff heavily bumping into their substrate diodes.

Regards, Joerg

 

[Adam S]

Hello Adam,


Strange, I have used the CD40106 as well as the 74HC14 in low power PWM
converters without any problems. They are both Schmitts, meaning they
won't dwell in the crossover phase any longer than their switching speed
lets them. They only consumed more when I went above a MHz or so but
that is to be expected from older CMOS processes.

But they do not like stuff heavily bumping into their substrate diodes.

Regards, Joerg

Yes, I would say the measured 600uA (@3.3V) IDD I got from the 74HC14
running at 500kHz is a 'low power' circuit, but not 'micropower'. An
LMC555 under same conditions drained about 200uA.
The gate output FETs are not the problem, its the input stage MOSFETs of
the Schmitt that suffer cross conduction. This is easily proved by
noting how current drain is a constant 600uA from 1Hz to several hundred
kilohertz.

 

[Adam S]

And maybe some provision to make sure -VOUT doesn't charge up to
+3.3v on startup, perhaps a schottky to GND?

As seductive as synchronous rectification is, in this instance the
inductor has ~8 volts flyback across it, so a schottky rectifier's 0.4v
drop will only contribute ~ 0.4v/8v = 5% overall efficiency loss. It's
pretty easy to lose that and more in switching losses in the sync.
rectifier, which already has an ohmic loss as well.

But, all these are moot: you've already got the thing beat just by
upping the LCD's divider resistances 10x. Touché.

The HD44780 based LCD character modules are actually capable of running
for a few months from just 2 AA cells.

As I said, modifying all the bias divider voltage resistors from
standard 1k to say 22k and by driving logic inputs of DB0-DB7, RS, R/W
to high state when idling so that HD44780 internal pull ups of these
pins do not add extra current, then these LCD character modules are
capable of consuming as little as 150uA@3V supply for the logic and
another 50uA@5V for the bias. Thats a big shift from 1mA@5V specified in
most applications.

 

[[email protected]]

Hi Joerg,

Hello Adam,


Strange, I have used the CD40106 as well as the 74HC14 in low power PWM
converters without any problems. They are both Schmitts, meaning they
won't dwell in the crossover phase any longer than their switching speed
lets them. They only consumed more when I went above a MHz or so but
that is to be expected from older CMOS processes.

But they do not like stuff heavily bumping into their substrate diodes.

That _is_ odd. Referring to my notes, the most recent time this bit
me was with a 74hc132 (quad 2-input NAND schmitt) 67Hz R-C oscillator.
Unloaded, it drew 700uA from a +4 volt supply.

That's been my experience right down the line with the HC-series
schmitts. I was possibly one of the first users of the 'HC14, or at
least would have been, except the early Nat. Semi and Motorola parts
dumped a HUGE, hysteresis-dependent bias current back at you through
the input! TI's parts did not do this but they came too late, so I had
to make my own (from 74hc244s).

Are you sure it was the 74hc14, as opposed to the 74c14?

Best,
James Arthur

 

[Didi]

Yes, I would say the measured 600uA (@3.3V) IDD I got from the 74HC14

running at 500kHz is a 'low power' circuit, but not 'micropower'. An
LMC555 under same conditions drained about 200uA.

Since you have found the 555 to be the lower power one, why don't
you make the next step - run it at 50 rather that 500 kHz. Over a 1 uF
capacitor this will yield something like 20 mV ripple p-p @ 1mA - is
this not good enough? The consumption should also drop dramatically
- may be not 10 times, but it will go low enough for >90% efficiency.
Well, if too much of the 200 uA go into the comparators you won't
gain much... but I would measure that before giving up.

Dimiter

 

[Joerg]

Hello Adam,

Yes, I would say the measured 600uA (@3.3V) IDD I got from the 74HC14
running at 500kHz is a 'low power' circuit, but not 'micropower'. An
LMC555 under same conditions drained about 200uA.
The gate output FETs are not the problem, its the input stage MOSFETs of
the Schmitt that suffer cross conduction. This is easily proved by
noting how current drain is a constant 600uA from 1Hz to several hundred
kilohertz.

Are you sure you haven't received some bad apples there? The Philips
spec says the typical added current if an input is held in the linear
region is 30uA. I guess that would be just one in your case. Even if you
held all six there that shouldn't result in 600uA. And this spec is for
5V, should be much lower at 3.3V.

Regards, Joerg

 

[Joerg]

Hello James,

That _is_ odd. Referring to my notes, the most recent time this bit
me was with a 74hc132 (quad 2-input NAND schmitt) 67Hz R-C oscillator.
Unloaded, it drew 700uA from a +4 volt supply.

Ouch, never had one that high.

That's been my experience right down the line with the HC-series
schmitts. I was possibly one of the first users of the 'HC14, or at
least would have been, except the early Nat. Semi and Motorola parts
dumped a HUGE, hysteresis-dependent bias current back at you through
the input! TI's parts did not do this but they came too late, so I had
to make my own (from 74hc244s).

Are you sure it was the 74hc14, as opposed to the 74c14?

Nope, never designed much with the C series. It wasn't mainstream enough
in terms of pricing and availability. Look at the Philips spec for the
74HC14, page 9, 5th line item:
http://www.semiconductors.philips.com/acrobat_download/datasheets/74HC_HCT14_3.pdf

30uA typical, and an oscillator doesn't spend a whole long time at that
very spot. Also, Adam mentioned 3.3V in which case it should be even
lower. It's usually good policy to run just one input on the RC node so
this stuff doesn't add up. Of course, if power mattered really badly one
could add a little MOSFET up front.

Regards, Joerg

 

[Winfield Hill]

Joerg wrote...

[ snip ]
Look at the Philips spec for the 74HC14, page 9, 5th line item:
http://www.semiconductors.philips.com/acrobat_download/datasheets/74HC_HCT14_3.pdf
30uA typical, and an oscillator doesn't spend a whole long time
at that very spot.

Not quite. Yes, once the threshold has been reached the switching
is fast and the extra current drain stops, but as the analog input
voltage slowly approaches the threshold the class-A current is high
and going higher, reaching a maximum just at the moment of switching.
See the graphs on page 14. So one gets a much higher supply-current
hit than expected if this hasn't been considered.

BTW, it's nice to see that parameter spec'd in the datasheet. But
one observes that a 30uA typical spec at Vcc -2.1V is not properly
at the tippy-top current peak! The stated 30uA pales in comparison
to the huge currents in figure 10 page 14, which peak at 330uA going
up and 420uA coming down. That's what we should be looking at IMHO.
The Philips graphs show that if you want 30 to 40uA current maximums
you'll have to run their 74hc14 at 2.0 volts supply! Hmm, too bad
they don't have any graphs between 2 and 4.5 volts, like 3.3 volts.

At any rate, I often end up avoiding the Schmitts for linear stuff.

 

[[email protected]]

Joerg wrote...

Strange, I have used the CD40106 as well as the 74HC14 in low
power PWM converters without any problems. They are both Schmitts,
meaning they won't dwell in the crossover phase any longer than
their switching speed lets them.

[ snip ]
Look at the Philips spec for the 74HC14, page 9, 5th line item:
http://www.semiconductors.philips.com/acrobat_download/datasheets/74HC_HCT14_3.pdf
30uA typical, and an oscillator doesn't spend a whole long time
at that very spot.

Not quite. Yes, once the threshold has been reached the switching
is fast and the extra current drain stops, but as the analog input
voltage slowly approaches the threshold the class-A current is high
and going higher, reaching a maximum just at the moment of switching.
See the graphs on page 14. So one gets a much higher supply-current
hit than expected if this hasn't been considered.

Yes, Figs. 10 and 11 illustrate the problem -- in operation, the
typical one-gate R-C Schmitt oscillator's input voltage range is
exactly the most disadvantageous vis-à-vis current consumption.

The 700uA-drawing circuit 74hc132 I mentioned before used a Motorola
74hc132 quad NAND-gate with R-C feedback to one input, and a
logic-level 'enable' signal applied to the other. All other
gate-sections were disabled. Dragging out the proto, I've re-measured
it.

It draws 700uA at Vdd=3.0v, not 4.0v as previously stated. At
Vdd=4.5v, the circuit draws 2.1mA.

.. 74hc132d
.. +------. +--+ Output:
.. enable+ >--------| --- \ | | 15mS pulse, 67Hz
.. | // |O----+-----> --+ +--
.. +---|--- / |
.. | +------' |
.. | |
.. | R1 |
.. | 2.2M |
.. o----/\/\/\--------o
.. | |
.. | R2 D1 |
.. | 68k In4148 |
.. +--/\/\/\---|<|----+
.. |
.. |
.. C1 -----
.. 22nF -----
.. |
.. |
.. ===
.. GND


Idd waveform @ Vdd=4.0v

.. _ ...__ 2.8mA
.. /| /|
.. / | / | rise/fall time = 11.4/1.6mS.
.. / | / |
.. / |/ |/_ ..__ 1.4mA
..

BTW, it's nice to see that parameter spec'd in the datasheet. But
one observes that a 30uA typical spec at Vcc -2.1V is not properly
at the tippy-top current peak! The stated 30uA pales in comparison
to the huge currents in figure 10 page 14, which peak at 330uA going
up and 420uA coming down. That's what we should be looking at IMHO.
The Philips graphs show that if you want 30 to 40uA current maximums
you'll have to run their 74hc14 at 2.0 volts supply! Hmm, too bad
they don't have any graphs between 2 and 4.5 volts, like 3.3 volts.

At any rate, I often end up avoiding the Schmitts for linear stuff.

Me too.

Regards,
James Arthur