Here is our circuit with a simple change: the base drive comes off the rectified LED output. This ensures that there is more than enough drive to saturate the transistor.   When power is applied, current flows from the battery through the diode, the resistor to the Base, starting the oscillation. After that, the LEDs act as a zener to maintain a relatively stable supply to the base.   The circuit requires a higher starting voltage to overcome the Vf of the diode, so it is useful to have a Schottky here. But once started, the circuit will run until the battery drops to Vce, which could be .3V or even less!   Another advantage to this circuit is that its output is fairly well regulated from 1 to 2 Volts and it is readily adaptable to 2- or even 3-cell circuits, because the base drive is less dependent on input voltage than other designs.
The circuit can even drive a 1-watt Lumiled at full output! Just remember the drain on the battery is over 1-Amp! A word of caution:- if the circuit does not start, INCREASE R! If the drive current is too high, the transistor will not switch off and the saturated coil becomes just a short circuit!
	By using a second transistor to act as a timer, the circuit on the left uses an off-the-shelf inductor, but is able to drive a 1-watt LED to full brightness with a 1.5-volt battery. Again, we can get much higher drive by utilising the 'bootstrap'.     The circuit operates at the maximum limits of the FJN965 transistor and the LED, so it is important to stay with the values given. You can use a 100-ohm resistor paralleled with a 470-ohm if you do not have the 82-ohm part.     Parts will run warm, but heat-sinking is only required for the LED - it gets as hot as 42C even then! And, if the paint on the inductor starts blistering, it's a darn good sign that you don't have one that can handle the 1-amp surges!
In early 2009, WatsonsEblog presented this Supercharged (SJT) circuit, which is explained in detail here and here     This simple and clever arrangement taps a bit of the output and feeds it back to drive the transistor. At the same time, the R2-C2 combination forces the circuit to run faster. As a result, the frequency goes up, but there is twice the power going to the LED.
	Like they say, anything worth doing is worth overdoing - how 'bout strapping on another engine? WatsonsEblog covered this, but he was using 2 transistors on the same Primary coil. What about making TWO Primary coils and powering each with its own transistor? Well, you'd end up with this circuit here. With one transistor, I measured 29mA going through the LEDs, and I added the other, and Presto! it doubled to over 50mA!     You'll notice the coil's value is quite low, that's because it's an air-core, easily made from 30turns of 3 wires, about 8-ft (2.5m) worth, wound over the 1/2" barrel of a large felt-tip marker. Mark each wire as P1, P2 and S1, use 28 or heavier wire for P1 and P2. Mark the Start of the wires with an "a"; P1a, P2a & S1a and make sure you wire the circuit with the proper wire connections!
These images show (1) the test circuit, with the big jumble of wire that is our coil, (2) running on just one transistor, bottom scale is the actual current into the LED, while the top is the power consumption and (3) with both on. The Red jumper on the left is plugged slightly differently.
THE FLASHING JOULE THIEF
An interesting fact about the 'Supercharged' circuit is that it starts at around 1v and doesn't turn off until the input is under 0.3v - just like a Schmitt trigger. So, by moving a few parts around we can turn this into a simple flasher!     As shown, the light will flash twice every second. The frequency of flash is approximately (1 / R1*C1) Hz and R1 can be any value from 400R to 5000R; while C1 can be from 100uF to 2000uF to allow for different flash rates (larger capacitor values will also make a brighter flash). The choice of a transistor is important here: the BC337 will work, but weakly. The best results I've had are from FJN965 (2SD965), 2SC2500 and, surprisingly, the MPS651, a medium power audio amplifier by Motorola. Also acceptable are MPSW01A and FPS530 1-watt amplifiers. Low performance transistors will likely not work.
	Taking an idea from the Turbo'd SJT, we can add an additional coil which is totally parasitic but gives us more than 10 times the output. Everything to the right of T1 is the basic SJT. The extra winding on the left side does nothing but supply the bias drive to the transistor, allowing us to reduce the size of the timing capacitor even further. Another advantage is that now the LED will be driven at maximum brightness for the entire cycle. Thanks to WatsonsEblog for the idea to put C on B+ to avoid any undershoot voltages. A short video is here on YouTube.     Details of the air-core coil can be found here, or you can substitute a toroid-wound inductor with 3 coils of 80-350uH per leg. The flash time is approximately R*C seconds, C can be 2uF to 30uF and you can adjust R from 47K to 1Meg. R2 is 1.5K for BC337 and FJN (2SD-) 965; 2.2k for PN2222 and 2N4401, and 1K for medium amplifiers like MPS651 and ZTX694. Use a 470pF for C1 if you are using 2 LEDs in parallel. The battery should be bypassed by a 5uF capacitor.     A more complete story of the development of this unique circuit can be found on the Ultimate Flasher page.
With a few extra parts, our circuit can be used as a very efficient Solar-powered flasher, ideal for upgrading Solar Garden lights. The inductor can be the air-core discussed, or a toroid with 3 windings of 80-300uH each. Light IS bright but will only draw about 15mA overall. Videos here and here.	Although the average power to the LED is under 20mA, surges are 4-5 times higher, so 1/2-watt (100mA) LEDs are recommended for longer life.     When the solar cell reaches about .5v above the battery voltage, D2 conducts and the battery begins charging. At the same time, the charge on C is drained through R1, keeping the flasher off. If, however, the battery reaches 1.6v, the flasher is turned back on - to prevent the battery from overcharging.     C can be 0.5uF to 25uF. R can be 100K to 1Meg. R1 should have the same value as R, but you can decrease by up to 25% for a later (darker) turn-on. There will be current leaking from B+ through the solar cell (it only has 20-50k internal resistance) so the time constant is approx (C * (R + R1)/4) seconds. Bypass the battery with a 2uF-10uF capacitor for longer battery life.     Change R2 to 2.2K for 2N4401 or PN2222; Try 1K (or less) for Medium Power transistors.
In fact, we can utilize the internal leakage of the Solar Cell in cheap garden lights to save ourselves another resistor (R) and we end up with this impossible looking circuit. The flash rate is now about (R1+20k)*C. I've had good results with a 3.3uF cap and a 150k resistor as R1.     The solar cell still works as described above, but, when it is dark, current flows from B+ through the solar cell, R1 and the P1 coil to charge C. When the charge exceeds the combined Vf of the diode D1 and Q1, it turns on and starts oscillations, lighting the LED. Each cycle drains a small amount of the charge from C until it is too low to sustain operations. The LED then turns off until C is recharged again. I've finally had to replace the NiCd from a garden flashing light which had corroded through and would not take a charge any more. This light had worked 24/7 through two winters. Hopefully a new battery will give it another 2 years' life!
	The 2-transistor 1-cell boost circuit can also be 'supercharged' for increased output. The parts shown here will allow a 1.2v rechargeable cell to light a 25mA white LED, change R1 to 68k and the LED will run at its absolute maximum. Care must be given because the power is delivered as a sequence of pulses, some as high as 100mA, but averages to 25mA. As a result, the light will seem brighter and more blue to the eyes.     The circuit will accept coils from a few uH to 3mH; C1 should be 5-10pF per uH of inductance, up to 0.05uF for 1mH and larger. In general, larger inductances will supply LESS current. R1 adjusts the brightness, and can be 22M-ohm down to 33K, which in this circuit will light 100mA and 1-watt LEDs.
	Just for completeness, here's a blinker for the dual-transistor booster. This causes truly eye-watering brightness, as each flash is almost 1/4 sec long and the pulses reach 250mA (Do not use low-power transistors for Q1 - they WILL burn out! Likewise for the LED!). The timing frequency is about 1 second per micro-Farad (uF) for C4 and C5.
Musicator JT - The Dancing Joule Thief How about a Joule Thief that dances to music? Running off the same 1.5-volt battery, this circuit uses an electret microphone to listen for sounds, amplifies it some 500 times and then uses it to modulate two LEDs, as seen on the videos below. Power usage is very modest at around 4 mA, peaking at 40-50mA for really loud passages. Many thanks to Botronics and AcmeFixer for their ideas. Transistors Q1, Q2 and Q3 form a Direct-connect 'gain-block'. The 250K trimmer-pot, TR1 maintains a DC bias on the base of Q1, but since all 5 transistors are direct connected, this voltage is used to control all the transistors, as well as compensate for differences in component values and LED voltages. The output of the electret mic is superimposed on this DC-bias to modulate the brightness of the LEDs.
SAMPLE RETROFITS
A simple installation driving 6-10mm LEDs (in 2 x 3-LED chains). These LEDs (available here on eBay) have a +/- spread of only 6 degrees, which makes them a good replacement for a flashlight - without needing a reflector. When driving multiple LEDs in series, care must be given to the transistor's Vceo and V rating for the output capacitor (especially tantalums). The peak voltage may be 50% higher than the measured voltage across the LEDs!
Another simple 6 LED setup with all the components mounted on a 1.5 inch square perfboard. Ultra-efficient 80%+ with 20mA into the LEDs and only a 300mA draw from an AA battery!	Next to a 3W Luxeon it has virtually the same throw and superior brightness, but uses less than 1/2-Watt!

	THE "HACK" CIRCUIT.
	This circuit uses the boost circuitry we have been using to lower (buck) a supply by placing it in series with the load (the LEDs) and feeding the "extra" voltage back to the smoothing capacitor.    This 'hack' of the boost circuit can be used as a buck regulator to run a reading light in the car - and it's 90% efficient! The voltage difference between the car battery and the LEDs, about 3 volts, is stored in the coil (seen as the flat part of the 'scope display), and periodically dumped back into the lights through the schottky diode (the sharp 14v spike). Or, use this concept to make a highly efficient flashlight for 9-volt batteries? Check out the HAK-LITE
This circuit is ideal for the latest Lithium- (and Potassium-) based batteries. As shown, the design will operate from 3 to 5 volts. The optional 1N914 small-signal diode acts as a very simple voltage regulator and lets us use this device for supplies up to 8 volts. With the 10k resistor chosen this circuit will drive a pair of 25mA LEDs in parallel between 15 and 30mA each.    In fact the circuit will work off virtually any NPN transistor; the output is only limited by the gain of the transistor, which can be offset by changing the value of the resistor (27K will halve the output while 6.8K will double it). I did a random sampling of transistors at hand, and found the 2N4401 and MPS651 solid performancers, and even the lowly PN2222 (the plastic version of 2N2222) managed 75% output.
Here are pictures of a Dollar-store "camping-light" which had the circuit inserted free-form through the (removed) bulb socket. It should give you 800 hours of use from a set of alkaline AA-cells. The LEDs used are 4.8mm 25mA bright-whites with a 120-degree coverage, available here. Click here for a detailled description on making your own.
	In this assembly, the LED wires are bent to form a trellis to support the 5uF capacitor and the other components "hanging" underneath. After everything is assembled (and tested), the coil and components are slipped through the neck of the light. The LEDs wires are then melted into the plastic rim by gently heating with a soldering iron. The inductor consists of two coils of 10-15 turns using 30-awg wire-wrap on a small (1/3" od) toroid, but anything from 200 to 1000uH can be used.
The simplicity of design lets us retrofit another 3-cell Camping light (from DealExtreme) and we can reuse the existing SPDT switch to control the 2 lights, each with different output levels!
THE UNIVERSAL HACK CIRCUIT
This same circuit, with the addition of a small timing capacitor, becomes virtually 100% self-regulating from 3-volt up to almost 20-volts, the Vce limit of the transistor we have chosen. The waveform across the transistor shows how it works: up to about 4-volts, the coil charges up through the LEDs, then, when Q1 shuts off, the coil discharges through D1, keeping the LEDs bright. Then the cycle repeats.   However, when the supply rises over 4-volts, the capacitor extends the time Q1 stays off in proportion to the input voltage, seen as the flat plateau on the second graph. This limits the current through the LEDs to a safe level.
This circuit is good for LEDs using up to 50mA; after that, we will have to look at regulated circuits, starting here.

---

Another take on the Solar Garden Light. With the addition of a solar cell, a simple diode, and some clever rearrangement of our circuit, we can devise a circuit which will charge a NiCd battery when it is light and automatically turn itself on after night falls. Can you explain how it works?
The MOSFET discussion section has been moved here!