Boost converter for low voltages

the technology behind energy harvesting

Similar projects worth following
While some like to look at the application side of energy harvesting, I like to look at the technology behind it.

Part of the problem of energy harvesting problem is to convert from low voltage source to a more usage voltage. I want to design such a converter using parts that I have and not rely on specialized chips or buy pre-built energy harvesting modules.

There are specialized energy harvesting IC on the market that does an excellent job.  A chip based solution works great if you need good efficiency at low working voltage, but each part are tailored for a particular range of operating voltage.

Energy sourceVoltage range
Thermoelectric20mV - 500mV
Solar80mV - 25V

 There was a conversation over here that illustrate the point for this project.  

The LTC3105 with an input range of 0.5V, can only source 5V around 22mA. The LTC3105 could be useful for bootstrapping a much higher power design. 

There are a lot of boost converters once the input voltage is above 0.7V.  

Where a gap in the market that isn't filled, you'll need to look for other solutions. A chip designed for energy harvesting has a lot of process advantages than conventional or discrete parts.  I still have a few tricks up my sleeves.

Transistor Basics

This is the $0.10 tour for a much more expensive/extensive Physic/Engineering course.   I am going to reverse the usual ordering.  This is about what I can remember these days.  :P 


N-MOSFET has a piece of P Substrate formed by doping with 2 N+ region (Drain and Source) formed by ion implantation (i.e. hitting it with a big electron gun).  There are back to back NPN diode junctions between the Source and the Drain which blocks current flowing between them.  The gate region is formed by depositing a thin layer of silicon oxide on top of the channel which forms a capacitor between the Gate and the Source.  

By applying a positive voltage at the gate, some of the free electrons in the substrate are attracted to the region and forming a negatively charged region called the depletion region.  This removes the diode junctions, so now electricity can flow between the Source and Drain.  A higher gate voltage increases the depth of the region and lower the resistance.

A large amount of current is needed to charge/discharge the capacitances quickly during switching to reduce switching loss.

The manufacturers play with the MOSFET structures to create devices with different electrical characteristics such as ones that can work with lower gate voltages.  A Depletion Mode MOSFET is formed with a depletion region by implanting N ions.  They turn on at 0V.  To turn them of, an external negate voltage is required to repel the electrons to collapse the depletion region.  The ones without this is called an Enhanced Mode MOSFET.

Bipolar Transistor 

A NPN transistor have a similar structure as a...

Read more »


LTSpice simulation file for "Depletion MOSFET LED driver from 0.05V"

x-zip-compressed - 13.17 kB - 07/15/2018 at 01:49


high current boost.asc

LTSpice simulation file for "Modding a boost converter for low voltage". Different PWM controller: LT3758, but same power stage as my mod.

asc - 3.85 kB - 07/15/2018 at 01:42


  • Depletion MOSFET LED driver from 0.05V

    K.C. Lee6 days ago 1 comment

    Here is a simulation of yet another LED driver of a much simpler design.  This time with a depletion mode MOSFET and only use 1 transformer.

    To get any power out of such a low voltage requires a lot of current, low resistance and a high turns ratio transformer.  

    The depletion mode MOSFET have a lower RDS(on) than JFET, so I can get away with a lower turns ratio.

    It is still not as good as a regular MOSFET that is wired in parallel to boost the current.

    This is the waveform at the gates of both MOSFET.  It is supposed to be a sine wave, but the top part is clipped by the LED forward voltage.  White LED (or blue) has a high forward voltage so the clipping doesn't affect the MOSFET too much.

    The depletion mode MOSFET has a negative threshold, so it conducts a bit longer and starts a bit sooner in the cycle.  Other than that both of them will be in phase.

    Blue Trace shows the inductor current while red trace shows the contribution from the second MOSFET.    Green trace shows LED current.

    The energy is transfer to the LED when the MOSFET are "on" similar to a forward converter.

    Another way is to reverse the LED polarity, so the LED turns on from the stored energy in the inductor core after the MOSFET are "off" (flyback converter)

    It would seem this has less drive as the inductor current is interrupted and allowed to decay in the middle of the cycle.


    Depletion MOSFET BSS159  \  They have spice models listed under their link. 
    MOSFET IRLM2402                 /

    BSS159 is used in the simulation as it comes with a vendor spice model.  Other ones such as Microchip LND01 could also work in theory.

    L1: Same one used in here winded on Pulse Engineering P0473 core.

    LED: NSPW500BS (Nichia) White LED (LTSpice built-in model)


    file: here

  • Modding a boost converter for low voltage

    K.C. Lee07/10/2018 at 18:15 0 comments

    I modded a LM3478 (3V to 40) eval board for low voltage boost.  With the modifications, I am hoping it would work for lower voltages once the output is running at 5V.

    There are a few changes:

    • RBYP is removed.  A schottky diode route VOUT to the VIN of U1.  This allow U1 to run from output.
    • Tabbed Inductor that allows for a higher boost ratio than a regular inductor due to duty cycle limits.
    • Removal of current sense resistor RSN and use Q1 for Ron sensing. (We can't afford 100mV drop for traditional current sense.)  2N7002 connects the drain to the Isen via 470R when the DR is high.  

    Part list

    Q1: I am using Si7806ADN (30V, 12A, 0.013R).  It is just a small N-MOSFET that I have - nothing spectacular.

    L1: Coilcraft DO3316T-223.  There are 22 turns on the core.  I made a tab at the 7th turn.  7T:14T

    This is the old version of the eval board when National Semiconductor was still its own company.

    To create small island of copper fill on the PCB for the mods, I use Polyimide tape (aka Kapton tape) for insulation and overlay small piece of copper tape.

    Load test:

    The bench supply was initially set to 3.5V for bootstrapping the circuit.  Once the output is running at 5V, the voltage is dialed back until it falls out of regulation. A bootstrap circuit could be used to provide this voltage instead.

    The lowest input voltage for this load is 0.45V.  My bench supply doesn't have enough resolution, so I hook up my DMM to read the input.  Smaller loads requires lower voltages.  

    The load is 70.2R drawing 5.05V at 72mA.  The efficiency is (5.05^2/70.2R)/(0.45*1.37) = 58.9%

    I can probably play with the inductor turns ratio and the MOSFET to try to get higher power.

    The following scope trace shows the gate drive to the MOSFET.


     I did a LTSpice simulation with a slightly different part: LT3758, but with same power components.  The simulation thinks it can supply 250mA, but that doesn't happen here.

    I made some changes trying to get more power, but it isn't possible.  The changes allows the same load to run from slightly lower voltage 0.376V.  It means it can run from a single solar cell under non-ideal condition.  

    • Q1: Si4842 (.pdf)  much improved MOSFET: 30V 24A RDS(on) = 0.0054R @4.5V
    • L1: 5.5T:16.5T
    • added a pull down a 470R at source of 2N7002, CSN changed to a 1nF.  Both of these are needed for the LTSpice simulation.
    • CIN1: Solid Aluminium capacitor 150uF
    • Thicker wires for input

    (5.05^2/70.2R)/(0.376*1.70) = 56.8%


    It turns out that the 1206 0R (Rsn) wasn't quite 0R.  I measured 40mV drop across the "0R" jumper, so there was some resistance there.  I have replaced it with a wide piece of copper foil and the boost converter now supports a 33R (36.2R) load (140mA) at 5.05V from a 0.449V source. The results now agrees with my LTSpice simulation.

    Efficiency = (5.05^2/36.2)/(0.449*2.31) = 67.9%  which is a big improvement from the sub 60% for only half the...

    Read more »

  • Lighting Blue LED from 70mV!

    K.C. Lee07/08/2018 at 00:23 0 comments

    I have wired up the following circuit to demonstrate energy harvesting from low voltage.

    The left side of the circuit is the JFET oscillator from JFET oscillator mod.  That mod increased the amplitude of the sine wave to 6.94V p-p which is sufficient to drive a small MOSFET IRLM2402 wired as a tabbed boost converter.  The transformer is a 1T: 120T transformer using the magnetic core from a Pulse Engineering P0473.  The high turns ratio allows a blue LED to light up from 0.07V.  The LED starts to light up dimly at around 50mV.

    I have built a low voltage supply for the 0.07V to supply enough current for the boost converter.  My regular bench supply have too much noise at this level.

    The following shows the AC waveform across the LED.

    The gate capacitance is wired in parallel of the ionizer inductor.  The flat top part of the waveform is clipped to the blue LED forward voltage.  I guess I have to watch out for those negative pulses frying LED. A small 30V schottky diode in series with the LED should protect it from the negative spikes.

  • JFET oscillator mod

    K.C. Lee06/30/2018 at 03:33 0 comments

    My JFET wasn't a particular good choice, but I was limited by the stock at my only local electronics store. It has a relatively low threshold which sets the negative feedback that limits the output swing of the oscillator.  Normally I would put a couple of resistors to raise the threshold, but the circuit couldn't drive any loads at all.  It can however handle quite a bit of pure capacitive loads as it merely shifts the resonant frequency without affect the Q factor too much.  I have tried up to 330pF.

    I made the following modifications to the circuit using capacitors divider which works like a voltage divider but only for AC.

    The threshold of the JFET is very loosely defined - it can vary from -0.3V to -3V which is an order of magnitude!  The gate capacitance of my JFET is 5pF (max) and there are some parasitic capacitance from the socket. All of these factors determine the right values to use.  I could have used a trim cap for the 12pF, but I got lucky with the first pair of capacitors I tried. 

    The two capacitors are just above the scope probe.  

    The oscillator now requires an input of 60mV instead of 22mV.  It is a small price to pay as the output went from 0.949V to 6.94Vpp

    A very large value resistor e.g. 10M connected in parallel of the 12pF would help to maintain a DC bias for the gate.

    I unwind the coils from a Pulse Engineering P0473.  There are 20 turns on the choke.  I measured 6 strands of 0.004" dia. enamel coated wire (AWG #38) at 14 in inches long and twisted them tightly together. It is more a logistical decision to make it easier to make a tighter wind.  Slightly thinner or thicker wires could be used, but this wire gauge seems to be just right for the number of turns to fit this core. I wind 20 turns to get to the same inductance of 880uH.

    I stripped off the enamel by tinning the end of the wire with hot solder.  The wire are thinner, so the windings have a much higher resistance which is still fine for what I need it for.

    I made a 1 turn primary turn with AWG 27.  Thicker AWG26 from CAT5 would be fine too.

  • JFET oscillator, how low can you go?

    K.C. Lee06/25/2018 at 20:46 0 comments

    I found Dick Kleijer's web site -- Oscillator with super low supply voltage.  I was curious to see how low I can get a JFET oscillator running with what I have on hand.

    The most tricky part was to find suitable transformer.  He was using audio transformers and a self made one for his experiments.  I bought some ozone generator last year hoping to use the circuits inside for a high voltage supply, but it turns out that the flash circuit inside a disposable camera works much better. 

    This is what's inside.  I figure that the transformer probably has at least 100X turns ratio as it was 12V in and probably a few kV out.

    Inductance L is proportional to (number of turns)^2. We can use this to figure out the turns ratio between two windings.

    I did some measurement of the inductances of the windings.
    The 4.5uH to the 400mH windings has a ratio of 3.4:1012  ~ 1:298 turns ratio 

    This is the wiring diagram (bottom view)

    Modern art 3D sculpture:

    This is the full schematic:

    The JFET I use is very old part for the analog front end of a project.  It is obsoleted and can probably be replace by any parts. There is a lot of voltage drop at the JFET as its nominal resistance is about 270 ohms. Dick is using several JFET in parallel probably to reduce this resistance and make the oscillator work at lower voltages.

    The JFET conducts when power is applied.  The current in the primary (3.4) winding increases gradually.  The secondary winding (1012) is connected the gate (G) of the JFET.  The transformer ratio steps up the very weak signal by about 300X.  As the negative voltage at the gate of the JFET reaches its threshold, it is turned off.  This cycle repeats itself and causes an oscillation at the resonant frequency of the transformer.   This frequency is determined by the parasitic capacitance and the inductance.  One potential use of this could be transmitting Morse code using CW from energy harvesting of thermal couples.

    The voltage divider on the right side provides 101X attenuation for my linear bench supply.  It makes life a bit easier to have fine voltage adjustments in mV scale.

    This is the output waveform between the probe point (Gate of JFET) and Ground at 22.4mV DC.

    The output is very weak and any attempts to load it causes the oscillation to stop.  With the components I have, this won't work for me.

    The problem is that the JFET (2N5484) I use has a much lower IDSS (vs J310 used by Dick.)

    May be something like the LND01 Depletion mode MOSFET might be able to squeeze a bit more power because of its much lower RDS(on).

    Spoiler: Depletion MOSFET can work!

  • Bootstrap oscillator

    K.C. Lee06/11/2018 at 18:29 0 comments

    I have decided to do some lab work as simulation can only get you so far when you don't even have the right model.

    I played around a bit trying to make my own inductor, a 15uH inductor with 4 windings.  I tried to make an oscillator, but I found that it "chirps" for a few cycles which dies down and could not substrain the oscillation.  The inductance is a bit too low as a result the oscillation frequency is too high for an audio frequency part.

    I have some Pulse Engineering P0473 Common Mode choke lying around and it is around the right inductance range I was looking for. 

    Note: As an EMI filter, common mode chokes are lossy at high frequency by design. I only care about low frequency.

    Here is the quick & dirty circuit I wired up based on what I had on the soldering bench.  I used a socket because Germanium transistors can be easily damaged by heat.  The resonant frequency should be around 300kHz (ignoring transistor parasitics, and +/- 20% tolerance on LC)  It can be lowered by increase the capacitance.  I'll need to play around with the component values.

    (Increasing the capacitor to 1nF increase the amplitude and the slew rate of the output.)

    The circuit starts oscillating at around 0.178V.  Here is the voltage waveform at the collect (ref to 0V)

    Here it is at 0.25V input.  This is measured at the feedback winding and 10K/330pF.

  • Simulation 2

    K.C. Lee06/06/2018 at 01:37 0 comments

    I made some progress by separating out the winding for the Germanium transistor feedback and the BJT/MOSFET pair.  This untangled the interactions of the bias setting and make tweaking a lot easier.

    V2 is added as a quick and dirty tweakable Zener clamp for the bias supply and keeps the oscillator stable.  If the bias is too high, the MOSFET won't turn off completely.

    I also separate the load and connect it (via M2) when the MOSFET is fully engaged. The staging helps the power supply start up.

    Right now the design supports a 3V supply at 30mA output (100R load).  I am hoping to tweak the design for higher power. 

    This shows the load sharing between the MOSFET and the BJT.

    Things are looking good.

  • Initial simulation

    K.C. Lee06/05/2018 at 02:15 0 comments

    This is what I have so far.  I do not have the spice model for the germanium transistor, so I used whatever I can get my hands on from the web.

    The circuit is similar to my LED driver, but with the 3 transistors are wired in parallel.  The germanium transistor Q2 helps with low voltage startup.

    The values of the bias resistors are a bit more complicated as they require some balancing with 3 sets of variables.  An additional bias supply from the output is used to increase the current via R5.

    Here are the plots for the current passing through the three transistors.  J-Fet J1 is used to cut off Q2 when the output bias supply after transistor Q1 starts to take over.  It is a bit crude due to the lack of parts that could work at these voltages.

    As the current further increase, MOSFET M1 starts to share the heavier loads.

    So it looks like the idea works to shift the heavier loads to the MOSFET. Here is a zoom in view of the current distributed between the 3 transistors.  I inverted the values for analysis

     The top graph shows the voltage drop across the transistors.  Looks like MOSFET reaches a saturation saturates around 2A.  Its gate voltage needs to be raise.  Q1 gets a peak current of around 400mA which is a bit high.  Both of these are related, so I'll need to adjust the bias level.

    This is a good start as the idea seems to work.  I am hoping the starting voltage would be a bit lower with the real circuit and may be some more refinement.

  • my junk/treasure box

    K.C. Lee06/04/2018 at 21:51 0 comments

    Found a few items in my junk/treasure box that are likely useful for this project.  God I love the transistor tester for these kind of things.

    Classic 2SB175 - Germanium low frequency low noise transistor from Matsushita Electric.  This is usually used for building distortion boxes these days.  This has a low Vbe, high Hfe and lowest leakage current in my collection.  This is probably late product before they are displaced by silicon BJT.  I don't think they do much spice simulation back then.

    Four 2N5484 J-FET in a bag.  Spice model also available on webpage.

    I think I have at least 1 UJT.  For now I think I'll be using the Germanium transistor and J-FET.

    I have a small solar panel from a solar lantern that has an output of 3V (open circuit) and current of about 26mA (short circuit current) in the summer morning sun.  It  is used to charge an Ultracap via a silicon rectifier diode.  The capacitor have very low impedance even near full discharged state which makes it an idea power source for the project.

    Q = C * V = 350F * 2.5V = 26mA * T   
    T = roughly 9.3 hours to go from 0V to 2.5V (too many variable to predict the exact time.)

    Result: It took 2 days to charge the capacitor from 0V to roughly 2V on my window silt.

  • Previous work

    K.C. Lee06/04/2018 at 20:50 0 comments

    In my Discrete Inverter for LED project, I explored the idea of using a MOSFET in parallel of a BJT to improve on the efficiency and to increase the output current for a boost converter.

    Here is a simplified version of the explanation..

    The BJT bootstrap the beginning of each conduction cycles.  The MOSFET takes over as soon as it gate voltage passes its threshold.  The voltage builds up at the secondary winding L2 due to a positive feedback as more and more current is allowed to pass into it.  When the inductor saturates, the voltage at L2 collapses and the MOSFET/BJT pair gets turned off.  The collapsing magnetic field causes a high voltage at L1 which is rectified by diode and stored in the capacitor C2.  Q2 is used for voltage regulation.

    The MOSFET reduces the voltage drop which is important for improving the efficiency for low voltage operation.  For even lower voltages, it is critical for good efficiency.

    There are two pieces of technologies:

    • Hybrid driver - use of the weaker part to generate gate drive for the stronger part
    • The bias voltage is derived from the output and that allows the circuit to continue working in lower voltage than its startup point.

    I am hoping to explore similar idea for this project. 

View all 10 project logs

Enjoy this project?



bobricius wrote 06/19/2018 at 20:25 point

nice work. Unfortunately all low voltage dc converters use special IC or historical transistors. This device look like work without IC and use SMD components

  Are you sure? yes | no

K.C. Lee wrote 06/19/2018 at 20:45 point


Or they would need to use depletion mode or really low threshold parts which I do not have on hand.

For me, making something moderately higher power than special chips sound like a reasonably original and useful application for the contest.  Germanium are still around and most importantly I have a whole bag.  I got to try or at least learn something with my failures.

There is also the trick of a momentary switch shorting the inductor that requires no special parts for start up.  Someone on 4hv told me about that.

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates