Overunity motor, part3, all 4 recharging bats reading at 1.400 volts now.

[Vortex1]

The real efficiency killer in a Joule Thief is its reliance on saturation of the magnetic core to switch.  The inductor current and circuit conduction losses shoot through the roof during the transition into saturation.  Using very square magnetic material as used in better magnetic amplifiers would reduce that loss a lot at a price of more expensive cores.  The $1. Chinese solar stick lights get around that by using a multivibrator circuit to time turning the switching transistor on and off.  They turn the transistor off well before the core approaches saturation avoiding the big uptick in current that occurs with an archetypical Joule Thief.  This also lets them use simple inductors that cost only a couple of pennies.  If you want to buy coupled chokes needed by a Joule Thief, they are around $0.40 in 5K quantities and $0.20 each in million piece quantities.  A number of those $1. retail stick lights get efficiencies close to 90%.  They are limited in the average power that they can deliver by the discontinuous conduction of the inductor.  10mA to 20mA average current to the LEDs is common.  They can be modified to continuously conduct to obtain more power, but at the cost of a substantial efficiency hit due to the need for output rectification and hard switching on both edges.

Agreed, for low power applications, the two transistor complementary multi-vibrator has many advantages over the single transistor blocking oscillator (aka Joule Thief) especially when it comes to controlling inductor current and avoiding saturation.
Blocking oscillators can be well behaved regarding core saturation, but that requires an extra active component to detect the onset of saturation and clamp off base drive or otherwise careful design of base drive parameters.....a level of circuit design savvy rarely seen in the JT community at large.

[MarkE]

Those little solar garden light circuits only use around 3mA ,and the LED apears quite bright.
Maybe a half bridge circuit that triggers at the 0 volt line across the driving coil.

The problem with the Joule Thief is the same thing that allows it to run at low voltage:  Timing by the transformer saturation.  The ICs that drive those stick lights are generally capable of delivering up to 20mA average without a great deal of trouble.  Some like the data sheet below can manage ~40mA down to a NiMH cell limit of about 1V.  At lower input voltages the efficiency isn't much better than a Joule Thief, but the required inductor only costs a few pennies and at 1V this one is generally better than 80%.  One of these would work:  http://www.digikey.com/product-detail/en/CBC3225T100MR/587-1625-2-ND/958009.  If you're looking for 25mA or so it's a pretty good circuit that costs less than $0.50 to put together.  Some LEDs will output around 10 lumens at 25mA, which is about the same as a 1W incandescent bulb.

There are other solutions out there as well.

[MarkE]

Agreed, for low power applications, the two transistor complementary multi-vibrator has many advantages over the single transistor blocking oscillator (aka Joule Thief) especially when it comes to controlling inductor current and avoiding saturation.
Blocking oscillators can be well behaved regarding core saturation, but that requires an extra active component to detect the onset of saturation and clamp off base drive or otherwise careful design of base drive parameters.....a level of circuit design savvy rarely seen in the JT community at large.

More transistors help.  Matched transistors help even more.  I've yet to see a Joule Thief or similar discrete circuit that could get more Watt hours to an LED load from a single cell: alkaline, NiMH, etc than one of the specialty ICs when applied properly.  The datasheet I posted just above does 85% - 90% efficiency over the entire run voltage range of a single NiMH.  With an alkaline the efficiency falls from around 85% at 1V down to around 70% at 0.8V.  There is only about another 5% energy in the cell from 0.8V down to 0.5V.  So a Joule Thief or other discrete design even if it performs as well as one of these ICs isn't going to wring out much more energy.  Since most won't come close to matching one of these ICs, KISS wins:  Go buy the IC from Digikey, and an appropriate inductor for $0.07 or so and call it a day.

[Vortex1]

More transistors help.  Matched transistors help even more.  I've yet to see a Joule Thief or similar discrete circuit that could get more Watt hours to an LED load from a single cell: alkaline, NiMH, etc than one of the specialty ICs when applied properly.  The datasheet I posted just above does 85% - 90% efficiency over the entire run voltage range of a single NiMH.  With an alkaline the efficiency falls from around 85% at 1V down to around 70% at 0.8V.  There is only about another 5% energy in the cell from 0.8V down to 0.5V.  So a Joule Thief or other discrete design even if it performs as well as one of these ICs isn't going to wring out much more energy.  Since most won't come close to matching one of these ICs, KISS wins:  Go buy the IC from Digikey, and an appropriate inductor for $0.07 or so and call it a day.

MarkE, you are correct, when it comes to getting the job done quickly and in a decent manner the IC will win. Unfortunately the proliferation and use of IC's by the experimenter unburdens the novice from having to learn discrete transistor design and perform it well. Thus you see so many in the JT community not really knowing how to read a transistor data sheet, let alone properly design a single transistor circuit. Most of the experimentation we see is cut and try, lacking circuit fundamentals and the discipline of real electronics design. It is no wonder most don't go beyond one transistor circuits.
    My experience spans three eras, tubes, transistors, and IC's. I still like to do discrete design that can sometimes match or exceed the performance of certain IC's', just for the fun of it, much the way some people like to do complex crossword puzzles, it keeps the mind sharp. Oftentimes it is a skill of limited value for today's high volume production runs which favor IC's, but great if your career choice includes analog IC design.

The Zetex part you mentioned is very nice for designs under 6 volts and low current. I only wish they had brought out the current sense resistor and a few other internals, but that's what keeps the cost down and application of the part easy. They probably have other parts with more flexibility.

[MarkE]

MarkE, you are correct, when it comes to getting the job done quickly and in a decent manner the IC will win. Unfortunately the proliferation and use of IC's by the experimenter unburdens the novice from having to learn discrete transistor design and perform it well. Thus you see so many in the JT community not really knowing how to read a transistor data sheet, let alone properly design a single transistor circuit. Most of the experimentation we see is cut and try, lacking circuit fundamentals and the discipline of real electronics design. It is no wonder most don't go beyond one transistor circuits.
    My experience spans three eras, tubes, transistors, and IC's. I still like to do discrete design that can sometimes match or exceed the performance of certain IC's', just for the fun of it, much the way some people like to do complex crossword puzzles, it keeps the mind sharp. Oftentimes it is a skill of limited value for today's high volume production runs which favor IC's, but great if your career choice includes analog IC design.

The Zetex part you mentioned is very nice for designs under 6 volts and low current. I only wish they had brought out the current sense resistor and a few other internals, but that's what keeps the cost down and application of the part easy. They probably have other parts with more flexibility.

There are other parts out there with more of the internal connections available.  I have a colleague who has some sub 1V designs that do mid 80's% efficiency from well under 100mW to over 300mW output.  The power transistors switch wicked fast which enables the circuit to run over 1MHz and still be very efficient hard switching.  For people who want to build a Joule Thief like circuit, unless the goal is to learn discrete analog design those Zetex parts are probably the way to go.