RustyBolt.Info/wordpress

I received a message asking several questions about the Joule Thief.  The asker asked to be explained in layman’s terms, so I’ll try to keep it understandable.  First off, I do not use a Spice program to simulate the circuit.  I used LTSpice a few years ago, but not any more.  Why should I when it takes a few minutes to connect five parts and I can then get the Real Thing?  Also, always keep in mind that a Spice program uses a component that is not like a Real World component.  A real world transistor might have a current gain of from 100 to 300, or even a wider spread, 100 to 500. I don’t think LTSpice gives you that kind of real world simulation.

Q1.  How did I measure the efficiency of a Joule Thief?

A1.  I used the supply voltage times the supply current to get input power (the regulated power supply is set for 1.5V and its meter tells me the current).  For example 1.5V times .075 A (or 75 milliamps) equals .1125 watts.  I measured the voltage across a 1 ohm current sense resistor in series with the LED. For example .017V times 1 ohm gives .017A.  I multiplied this current times 3.3V for the LED forward voltage to get the output power to the LED. For example .017 times 3.3 equals .0561 watts.  Then I divided the output power by the input power to get the efficiency. For example .0561 divided by .1125 equals .4987. Multiply this by 100 percent gives 49.87 percent efficiency.

Q2.  Do you have any recommendations on measuring efficiency, with a DMM, without a regulated power supply?

A2.  You can measure the battery voltage with a DMM, at the point where the battery connects to the circuit.  Another 1 ohm resistor can be used in series with the battery to allow the DMM to measure the voltage across it and then get the current, same as above example.  We have to remember that there is about 100 mA supply current, and that means about 100 millivolts or 1/10 volt drop across this 1 ohm resistor.  That’s about 7 percent of the 1.5V supply, and it’s too  much V drop.  We can lower this by using a lower value resistor.  A 1/2 ohm resistor will give about 50 millivolts drop, which is better.  I used 1/10 ohm resistors, which give only 10 millivolts drop.  That’s less than 0.7 percent of the supply voltage.

Q3.  I know that the 2N3904, 2N2222 and 2N4401 are all different types of NPN transistors but what particular characteristic makes one more efficient over the other (in this application)?

A3.  The transistor should have high current gain at high currents and very low collector voltages.  A maximum collector current of 500 milliamps is a good starting point, the more the better.  It must also have a low Vce(sat) at high currents. I look at the graph and see if the transistor’s collector saturation voltage is 1/4 volt or less at currents of at least 250 mA, but hopefully at closer to a half amp.  The BC337 does not have any graphs for Vce(sat) but every BC337 that I have used can handle current better than a PN2222A or 2N4401, so I recommend it for JT use.

Q4.  You also mentioned the JT eval website on scribd, so you may or may not be able to answer these questions but: http://www.scribd.com/doc/35050054/Joule-Thief-Eval  1.  What exactly is a figure of merit? Is this just another name for efficiency?

A.  I cover the stuff in this document in my recent blog.

Q5.  How does the capacitor (base capacitor from the JT eval) allow the transistor to switch on and off faster? What are the consequences of this and why?

A5.  The author explained this in the document at the link you gave.  But I have my opinion about that, and it is covered in my recent blog (see the link above).

Thank you for the thoughtful and thought provoking questions.  I hope my answers are correct, or if not, reflect what results others may get.  As with the other author, his mileage varied because of his situation.  Your mileage may differ if you do the same experiment.

I often read where people add the rectifier and filter capacitor between the coil/transistor and the LED.  They claim that it increases the efficiency.  I just want to clarify what I see is happening as the Joule Thief is operating.  My observations are based on what I see on an oscilloscope connected to the JT, and other measurements.

In a JT, the coil stores the energy that has to go somewhere when the transistor turns off, and this gets transferred to the LED.  During this time the transistor is open, it as if it was disconnected, and it does not dump the energy to ground. If you add the Schottky diode and filter capacitor, the schottky does not increase the amount of energy in each pulse that comes from the coil.

When your JT is running, each pulse of energy from the coil starts at a low voltage and high current and gets transferred as a higher voltage at a lower current.  To my thinking it’s a pulse of energy, and whether you transfer it directly into the LED or through the diode and filter capacitor, which turns the pulses into filtered direct current, the energy from the coil is the same.  But the diode has a voltage drop and when you find the power by multiplying the current times this voltage drop, this is a loss of power that ends up heating the diode – it is wasted power.

There may be other factors, such as how the LED responds to pulsed DC and filtered DC.  The filtered DC could be brighter because of something that is caused by the LED’s response. But the way I see it is the LED is already a diode, and the circuit does not have to have a rectifier and filter capacitor to make it more efficient.

In my recent blog, I compared the conventional JT with my Supercharged JT, and gave the results of the actual light output measured with the luxmeter.  The SJT is much more efficient, and is a simple circuit using inexpensive parts.

One might ask why I don’t compare the light output of the SJT with a conventional JT with Schottky and filter capacitor.  I can’t see why I would need to because I believe the SJT has to be more efficient because of the lower current and therefore lower loss in the (Schottky or regular) diode.

Update Jan 6 2012 Comparison W & W/O Schottky & Filter Cap

I used a slide switch to switch the LED between the conventional (without) and the 1N5817 Schottky diode and filter capacitor.  First, the components I used.  The coil was a 1/2 inch high mu core with 13 turns solid twisted pair from a cat5 cable.  The LED was blue, with a 1 ohm current sensing resistor in series.  The transistor was a BC337-25.  The filter capacitor was a 100 uF, and the supply had a 100uF bypass cap.  The supply voltage was 1.5V for both tests.

The LED current was 19.7 mA and the supply current was about 95 mA without Schottky.  When I switched to the Schottky and cap, the LED current went up to 24.7 mA, and the supply current went up to about 98 mA. This is a substantial gain in brightness with only a small increase in supply current.  This agrees with the claims by others that the Schottky diode and filter cap increase the brightness of the LED.

However I must remind everyone that my Supercharged Joule Thief does even better with easier to get parts and less expense.

Update Jan 10 2012 – QS left a comment; Thank you.  I’m happy to see he is somewhere, because Bill and I were seeing his emails bounce because his email inbox was full.

Joule Thief Special  There are thousands of different types of transistors, made by hundreds of manufacturers.  But it takes a special kind of transistor to give good performance in a Joule Thief.   The transistor should have high current gain at high currents and very low collector voltages.  A maximum collector current of 500 milliamps is a good starting point, the more the better.  Most transistors have a loss in current gain as the current increases past a certain point; for small signal transistors that point is about 100 mA or 1/10 of an amp.   The datasheet for the (Vishay) BC337 shows the current gain beginning to roll off before 100 mA, but by the time the current gets to a few hundred mA, the gain is still 200, so it still makes a good JT transistor.  The 2N3904 datasheet (Motorola or On Semi) graph shows the gain at 10 mA is around 240, but at 200 mA, it has dropped drastically to about 30.  This is a poor choice for a JT.

The JT transistor must also have a low Vce(sat) (collector to emitter saturation voltage) at high currents. I look at the graph and see if the transistor’s collector saturation voltage is 1/4 volt or less at currents of at least 250 mA, but hopefully at closer to a half amp.  The BC337 (Vishay) shows the Vce(sat) rising above 0.25V when the collector current gets up to more than a half amp.  Every BC337 that I have used can handle current better than a PN2222A or 2N4401, so I recommend it for JT use.  The same Motorola datasheet for the 2N3904 shows that the Vce(sat) rises above 0.25V when the collector current is about 75 mA, and at 200 mA it is over a half volt, and that is very bad, because it is wasted power heating the transistor and not lighting the LED.

The following is a table of transistors used in the same JT, and the supply and LED currents that I measured.  The 2N3904 is very close to the BC550 performance.

Recommended:

2N4401, PN2222A, BC337, BC338, SS8050

Not recommended:

BC547 series, 2N3904 (the JT exceeds its maximum collector current of 200 mA.)

Happy New Year!  My watsonseblog dated 2009 Jan 1

Joule Thief, FET Only is the topic. In April 2005 I built a Joule Thief (circuit is in my blog and at the bottom of this webpage) that used both a 2N3904 BJT and a 2N7000 MOSFET, with the 2N3904 starting the circuit and the 2N7000 then supplying most of the current. It’s not a complex circuit, but it really doesn’t need the 2N3904 after it gets started and the 3.3 volts across the LED is then used to supply the gate bias to the 2N7000. The 2N3904 won’t start at a supply voltage less than 0.6V, but it could be replaced with a germanium transistor which operates easily down to 0.2V. But germanium transistors are getting very rare, and they are very temperature sensitive, which is probably why silicon transistors have replaced them almost completely.

Theoretically Speaking..
If a MOSFET had a gate that would operate below a few volts, then it alone could be used for a Joule Thief or any other low voltage circuit from a single 1.5V cell. But the gate of almost all MOSFETs requires more than 1.5V to turn it on.. With a Joule Thief, though, once it gets operating, the boosted and rectified voltage can be used to supply the gate bias.

And remember that this is a MOSFET, which means the gate is insulated and does not draw any current from the circuit. The base of a BJT, on the other hand, only operates with a small current flow. If we can raise the voltage at a MOSFET’s gate to turn the MOSFET on, there will not be any current flowing thru the gate. The feedback will have to charge and discharge the gate to source capacitance as the gate’s voltage changes, but this capacitance is almost purely reactive, with almost no resistance, so it doesn’t dissipate any power.

Thus, I came up with the idea that I could make the MOSFET work at very low voltages by putting a voltage source in series with the gate, so that it is “prebiased” closer to being on when the circuit is powered up. I used a 1.5V alkaline button cell to add 1.5V to the gate when the circuit is not powered up, and when the 1.5V power is applied, the gate will have a total of 3V on it, which is enough to get the MOSFET to run by itself. Since the gate draws no current, the button cell should last for as long as its shelf life, which could be several years.

I read about the Zero gate threshold MOSFET. I downloaded the spec sheet for the ALD zero gate threshold ones, and found that the Rds(on) is 500 ohms at max (but probably a few hundred typical). What this means is that at a fraction of a volt, the most they will conduct is a milliamp, so if they are used in a converter, they will only put out a fraction of a milliwatt. But that’s okay, one would use this MOSFET to boost up the voltage to 5 or more V, then use that at very low current to bias the gate of the power MOSFET.

So I did some more searching for a better solution and came up with the Supertex line, specifically the TN0702N3, which is $.63 each if you buy 100. They are low gate threshold V, 0.5 to 1V, which isn’t as good as the ALD parts, but a lot better than the 2N7000. The Rds(on) is a lot lower than the 2N7000’s 5 ohms. I ordered them from Mouser, along with some button cell holders.

The Circuit
I built the circuit shown at the top in the schematic, and I used a coin cell holder to hold the much smaller button cell. It is much too big, but the holder’s spring holds the button cell good enough to make contact. I put a 100k resistor in series with the button cell’s + lead, and I measured the voltage across it while the circuit was operating. On the lowest V range, the voltage read 0.0V, meaning that there was no current flowing through the button cell.

The T1 toroid has to be wound different than the usual Joule Thief. The feedback winding has to put out a higher voltage to the gate, at the gate’s very high impedance. This means the feedback winding has to have many more turns than the primary winding. I didn’t count the turns, but the feedback winding has ten times the inductance of the primary, so it has about 3 or 4 times the number of turns. The feedback winding should be able to develop enough voltage to cancel the 3VDC that is on the gate and turn the MOSFET off. I don’t know if this is enough turns, it might work better with 5 times or ten times as many turns. That’s another area of experimentation that I’ll have to try.

RESULTS
As it is, it works very well. With 1.5VDC supply, the circuit draws about 80mA from the supply, puts out 13.2mA to the LED, and runs at 71kHz. I kept lowering the supply voltage until the LED went out at below 0.4V. It also started up at 0.4V, which is better than a silicon BJT transistor, but not quite as low as a germanium BJT. But this may be changed by increasing the button cell voltage and/or increasing the number of turns in the feedback winding.

Update — a few hours later. I wound another two feet of 32 AWG wire onto the core and connected this and the original feedback winding in series aiding. (the aiding is important — if they’re opposing, one will cancel out the inductance of the other, and the total will be less than either winding). Their total inductance is 2.8 millihenrys. I’m estimating that the primary-feedback turns ratio is about 1 to 5 or a bit more. Now my ‘MOSFET Joule Thief” has increased LED current to 15mA, supply current to 110mA, and a frequency of about 65kHz. But the really cool part is the circuit will now start up at below 0.3V, around 0.27 volts supply voltage. This is getting very close to what the germanium transistors will do. And at 0.5V, where the regular transistor can barely light the LED, this circuit can still put out a substantial amount of light, much more than the standard BJT circuit, in my experience.

Further results..
I removed one turn from the primary, making it ten turns. The supply current went down a few mA, the LED current went down to 14.7mA, and the frequency went up a bit to 69kHz. It starts up at about 0.27 volts, not much difference from before. Even though the turns ratio got greater, the LED current went down, so I think I’m heading in the wrong direction.

I put a second MOSFET in parallel with the original 2N7000, and it made a significant difference. The supply current went up to 175mA, the LED current went up to 19.7mA, and the frequency was 60kHz. The increase in the LED current was welcome, but it came at a much higher supply current and a lower efficiency.

I put a short across the 0.1uF capacitor and the LED went out immediately (the 100k resistor is between the cell and the capacitor, so I’m not shorting the cell). This shows that the additional 1.5V is needed for the circuit to continue to run.

More Changes
I did some more experimenting with the circuit. I added a second button cell for 3V, and put a 470k trimpot across them, so I can adjust the voltage going to the 0.1 cap and then to the gate. I changed the gate bias voltage with the pot while monitoring the supply and LED currents. As I increased bias, the supply current increased a lot but the LED current went up only slightly. So I’m guessing that the on time of the FET is increasing, and letting the current go to waste. So I don’t think the bias voltage needs to be more than 1.5V. In fact, I can turn the pot down and the circuit will work with less than a volt from the button cells. I’ll be writing this up in my blog asap. I figure that the 470k pot’s drain on the button cells might discharge them in a year or so… 😛

I’ve still been playing around – er, experimenting with my circuit. I removed turns from the feedback winding. It was 5 or 6 times the turns of the primary, but now it’s down to about 2 times. The frequency went up from 60 or so kHz to about 100kHz, and the LED and supply currents have come down, but to increase them, I turned up the bias adjust pot. The LED current got up to 17mA, then when I went further, the supply current jumped to over 200mA and the LED current peaked at 20mA. I wonder why the thing jumps to so high a supply current (luckily my supply is current limited). I added back a few turns to the feedback winding and now the current jump went away. It might be that the gate bias is taking the MOSFET into the linear region, where it is conducting heavily but still oscillates.

Running on Empty…
Just for the halibut, I decided I wanted to find out if the 0.1 uF capacitor could hold the charge of the button cells by itself, and for how long. I disconnected the button cells from the capacitor, leaving only the capacitor, the feedback coil and the MOSFET’s gate connected, and the wiring between them. The LED continued to stay lit. I monitored the LED current through the 1 ohm resistor; it started out at 13.6 milliamps. As time went on, it dropped a tenth of a milliamp, then another tenth of a milliamp. After about an hour, it had dropped about 2 tenths and continued to drop. Six hours later it was down to 11.2 milliamps. It seemed to drop about a tenth of a mA every 20 or 30 minutes, so the capacitor is holding the charge very well. Unfortunately I can’t measure the voltage direcly because as soon as the meter touches the capacitor, it discharges through the meter!

Button Cell Holders?
I think I need to buy some 1 cell and 2-cell button cell holders. I’ve never seen one in any equipment; every button cell I’ve seen has been in a holder made into the plastic case of equipment. Maybe Google might be able to help. In my blog I previously described one I made out of a wood clothespin, which works okay but is much too big for a small MOSFET Joule Thief circuit. I’ve also described others I’ve made in my blog.

Well, I found some button cell holders here. Not quite what I expected, though, just a metal clip. But I found something really cool there too. Put your mouse cursor over the picture of the button cell holder and see what I mean. Cool! I don’t think I’ve ever seen that before. I’ll keep looking for something more substantial – besides the company sells the holder 30 to a bag, and like I’ll probably need a dozen in the next decade, I’d really like to find a smaller quantity. Not to mention that the company is halfway around the world, in England. Well, I found some 12mm button cell holders in the Mouser paper catalog, so I’ll have to order a few, and some white Seoul Semi LEDs while I’m at it. 😉

I ordered some 12mm button cell holders, both single and dual, from Mouser, $.99 each, USD. Oops, I already have some Seoul Semi 1W LEDs, so instead I ordered some Supertex N-channel, Enh mode MOSFETs, TO-92 pkg, low threshold, 0.5 to 1V. Cool! Mouser # 689-TN0702N3 bulk packaged, catalog P. 523. And a few Fairchild J105 JFETs, which are supposed to be able to handle 500mA. They’re used in the startup circuit in the Damaschke .PDF (google ‘damaschke thermopile’) about building a DC-DC converter that runs off a 0.3V thermopile.

Other Thoughts..
Another thought occurred to me. If you wanted to, you could make a “Joule Thief Squared”. What I mean is that the button cell could be replaced by a depleted AA or AAA cell, or even two depleted cells in series. Or even spent button or coin cells These would last until they dried up or leaked, since there is no current flowing through them. Just another way of stealing more Joules from spent cells – that’s what it’s all about. (This reminds me of my old calculators and I’ll have to spend some time writing a blog on that subject.)

Another thought: If we can make an electret microphone that uses a “battery” made by exposing a substance to an electric field, we should be able to make an electret with enough charge to put in a MOSFET circuit so that the gate would need much less voltage. Just like the button cell, it would add to the incoming signal to allow the incoming signal to turn the gate on at a fraction of a volt. In a way, this would be like a depletion mode MOSFET, however, the enhancement mode MOSFET would not be conducting with zero bias on the gate, as a depletion mode MOSFET does.

Another thought. If I could get the insulation so high that there would be almost no leakage, I could put a capacitor in series with the MOSFET’s gate, and charge the capacitor up to the required voltage to bias the gate just below turnon. Then with a signal of just a fraction of a volt, the MOSFET would be turned on. Since there is no current flowing in the gate, the capacitor would stay charged for a very long period of time — maybe months or years if the insulation was good enough. If a ‘super capacitor’ of, say, 1 Farad could be charged up and hold its charge for years, then it could be used in this circuit. The advantage here is that the voltage could be adjusted, in contrast to a battery, which is always at a fixed voltage.

I did some google searching and found that aldinc.com makes a zero gate threshold MOSFET, and it will work to below 0.2VDC. See this link for a circuit example. http://www.discovercircuits.com/H-Corner/verylowosc.htm Wow! Cool! You don’t even need a button cell!

I’m going to go back and rewind the toroid, or wind more turns in the feedback winding. Well, I just did that, and updated it above, with definitely positive results. Now the hole in the toroid is just about filled up. If I want to add more turns, I’ll have to get a bigger core, or use finer wire. I may remove a turn or two from the primary, which will increase the turns ratio, but the frequency will go up. That’s not a problem for the MOSFET. I’m wondering when I’m going to meet the point of diminishing returns.

Ideas From Others..
I corresponded by email with Quantsuff, and he came up with the nifty idea of using a DPDT switch to turn the circuit on and off, but also switching the 0.1uF capacitor from in parallel with the supply (to charge it up) to in series with the supply, thus giving the MOSFET gate a total of 3V at startup. Then after it gets going, the output to the LED is rectified by a diode, through a resistor, and back to the 0.1uF capacitor to keep it charged. It will then have the LED voltage minus the diode drop across it during operation, which is about 3 volts.

This is great, it eliminates the button cell. But one problem is the circuit always needs human intervention to start up. Take a solar powered yard light for example. At sunset, you would have to go to every one to turn them all on. Another concern is as the “MOSFET Joule Thief” runs down the battery, the bias voltage across the 0.1uF capacitor also goes down (from diode leakage), so I’d guess that this will adversely affect the performance.

This “flying capacitor” or “switched capacitor” charge pump circuit is used by many DC-DC converters, but it is switched electrically, not manually. National’s LM2751 and Maxim’s MAX1576 are two of many ICs that use this. Dave Johnson has a switched capacitor LED driver here. The idea is to double the voltage by charging the capacitor then putting it in series with the supply. With a similar circuit, one could draw a few microamps from a single 1.5V button cell and double, triple or quadruple the 1.5V at very low current to make enough voltage to bias the gate of the MOSFET.

In his email, Bill “Botronics” Sherman used a small solar cell in place of the button cell. That’s cool. But of course the circuit won’t start up in the dark. But then, this might have an advantage: you could start it up by shining a beam of light at the solar cell from a distance. The solar cell costs five bucks, so I don’t think that it would be cost effective to use one to turn off a circuit that cost 1/5th that much.

Back to experimenting…

My watsonseblog dated 2009 Feb 19   These may be dupes or unfinished.

2009-02-19

Joule Thief, High Efficiency, #01 Supercharged

See the schematic and original blog here.

I have built several of these high efficiency ‘supercharged’ Joule Thiefs (SJTs) to try to optimize the parts values, especially the coil. I’m posting these SJT blogs to allow the reader to get better results when building a duplicate SJT. I’ll give info on winding the coil, wire size, etc.

This SJT used a toroid core from All Electronics, part number TOR-54 (they have since sold out). I put a single turn (piece of straight wire) through the hole in this core and it measured 0.08 uH (I measured several of them to get an average). This means it is a low permeability core, much lower than the usual ferrite RFI suppressor sleeves that I’ve used (compare that to the charcoal TOR-23 cores they sell, which have 0.48 uH per turn). It also means that I have to use more turns to get the inductance up to 100 uH.

I jumble wound three solid enameled wires, each 36 inches long, wound trifilar, i.e. all three at the same time. Two were 28 AWG, connected in parallel for the primary winding. The third was 30 AWG, for the feedback winding. I cut the three wires to length, and on one end I twisted all three together to allow me to thread the wires through the core. Afterwards I cut off the twisted part. Each winding measured 124 uH.

For Q1 in the Fig. 2 circuit shown in the schematic (see link at top), I used a BC337-25. The R2 was 820 ohms, D1 was 1N4148. C2 was 1000 pF. The LED was a cheapo white LED removed from a 9-LED flashlight. A 47uF tantalum was across the supply rails.

MEASUREMENTS
I connected it to a 1.5V supply, and measured the supply current at 50mA. The LED current was 21 mA. The freq was 224 kHz. The LED was very bright. I calculated the efficiency at 89 percent, which seems too high, however the circuit did have very low power consumption.

Observations
This coil was wound using .335mm wire for the two primary windings. This is actually slightly larger than 28 AWG, about 27.5 AWG. It’s taken from a fan motor winding made in Switzerland. I had to scrape the very tough insulation off the ends with a single-edged razor blade, and it took quite a bit of scraping.

I am trying to keep the coil inductance above 100uH, and with the TOR-54 this means using solid enameled wire. If I used plastic insulated telephone wire, I wouldn’t be able to get a long enough length on the core to get 100uH.

Back to experimenting…