Efficiently dim your LED without a sense resistor

When driving a single LED from a battery, efficiency is key.  Efficiently making use of the battery’s energy and efficiently converting it to light output prolongs battery life, which reduces cost and hassle for the end user.  While high efficiency switching power conversion is the first step in this process, dimming is a second step just as important as the first.  Dimming allows the user to reduce the light output, and thus the battery power consumed, when they don’t need such a bright light.  This combination of dimming and high efficiency power conversion gives such items as smartphones and glucose meters a sufficiently long battery life.

Comparison of LED current regulation through a) a sense resistor and b) a MOSFET in the power stage.

A further step to make these devices even more efficient at driving an LED is to improve the efficiency of the switching power stage itself.  An important loss term in every LED driver is the loss in the current sensing element—typically a sense resistor.  Being a resistor, all current going through it is a loss and a drain on the battery.  In order to achieve a tight control over the LED current (which is proportional to the light output), a resistor is usually used as the sensing element.  The LED’s current is routed through this resistor and the voltage developed is measured and regulated by the LED driver.  While simple, this is unfortunately a relatively large loss for a single LED driver.

In most switching power supplies, there is a MOSFET in the power stage that can be used to sense the LED current instead of a resistor.  This method requires more processing of the current signal, but when integrated inside the LED driver IC, this complexity is removed from the user.  Very simply, higher efficiency results.  And the larger and possibly costly sense resistor is no longer needed in the design!  As an example, I've included a comparison above of the input current drawn by the TPS61260 LED driver when using either a sense resistor or MOSFET to sense the LED current.  A full analysis of this design is here (LEDs Magazine might require a login for this) and it can be tested here.  Let me know if you have any questions?

Anonymous
Parents
  • Chris,

    It is a great thought.

    One flaw that I feel is a pretty big one, which can be bit difficult to deal with- one that used a similar method for current sense by using the MOSFET for the current limit in the TPS40071/77 family.

    First, the accuracy of the internal reference in the chip from part to part, and the offsets of it's internal amplifiers- Notice how the TPS61260 datasheet is devoid of current accuracy information over temperature, and missing most of the important information an engineer should be considering - these sorts of omissions from datasheets usually indicate a huge red flag- an area that one needs to carefully consider and test carefully.

    At the currents you are showing, a flashlight would be a good application.  Realistically, the temperature of my flashlight can range from 7 degrees F and if it is warm out (say up in an attic on a hot summer day doing an inspection) it could be up to 120 degrees F ambient, and if operating for 15 minutes, internally it could reach 160 degrees F- easily.  This range could be much wider, depending on where in the world you are, and time of day.

    Since the part you mention is devoid of critical details, lets use the TPS40071:

    Internal Error amp feedback regulation voltage low of 0.690V and up to 0.715V

    Input bias current -250nA to 0nA (creates an effective error in the feedback divider)

    Skip to current limit offset voltage -75mV to -30mV over temperature.

    Add to that the current sink into the current limit, which varies from 80 to 125uA.

    Now, lets jump to an external MOSFET one may use, to give us an idea of what happens to the MOSFET, especially considering temperature. Lets use a common MOSFET datasheet, say the FDS6612A found here:

    www.fairchildsemi.com/.../FDS6612A.pdf

    In figure 3, you can see the temperature causes the on resistance to vary by a factor (1 being the value at 25C).  7 degrees F is -14 degrees C, and 160 degrees F is 71 degrees C.  However the die inside can be much hotter, and there isn't much room in a flashlight, and I've never seen the MOSFETs drain, or "thermal pad" attached to the body of the flashlight.  The boards are tiny, and they don't have room for even the minimum pad.  So lets say the die is cold at first, and eventually internally it heats up an additional 50 degrees C over 15 minutes.  So, we are looking at -14°C to 121°C (71 +50). So, using Fig. 3, we see the MOSFETs' on-resistance will vary from a factor of more than 0.9 to 1.4, about a 50% variation in on-resistance (depending on how you look at it).

    Now, on the same datasheet we jump back to page 2, and look at the Rds(on) which is specified at 25°C, lets use the 10V gate drive, nominal 0.019 ohms to 0.022 ohms max, with no mention of the minimum. From experience, the minimum can easily be the same amount lower, so lets use 0.016 ohms for that.  Remember, this is at 25°C, so we already get a 28% variation part to part, add to that the 50% percent on-resistance variation due to temperature, and you are looking at a 78% variation of the MOSFET on-resistance over temperature.

    Now, go back to all the errors accumulated in the chip's reference, bias current, and current limit offset and consider it all together.

    It gets even tougher if you need the current limit to always function as it should, and you need to specify in a MOSFET that can handle all the errors, without letting out the smoke- seriously increasing the cost of the design, and lowering the efficiency.

    It would be nice if the chip designers would consider these basic realities, and compensate for them.  However the designer probably figured the chip would end up in a cable/dish control box that sit's in a climate controlled living room, and assumes the designer has plenty enough room to put a 1" square patch of 2 oz. copper on the MOSFET, or has room a heatsink for the SO-8.  Although this isn't the case much of the time, were one may be limited by space, no flowing air, no heatsink, and has to use 1/2 oz copper due to a part that is only in a CSP scale package, or 0.3mm pitch BGA FPGA.

    Though the idea might work okay for a really low power flashlight that uses a T1-3/4 LED, but their price point is already so low that they use the button cell's own internal resistance to limit the current, since they can't afford the cost of all the additional electronics.  Though maybe it could make sense in a cell phone.

    Consider all this again when you have 350mA to 3.0 Amps flowing in the LED in common flashlights where the price point is high enough to have decent electronics.  Some of the high performance flashlights are driving 9 Amps into a Luminous Devices LED, or using an array of multiple high power LEDs from companies like CREE, Royal Philips LumiLEDs, OSRAM, etc...

    Just some thoughts to muse/chew on.

Comment
  • Chris,

    It is a great thought.

    One flaw that I feel is a pretty big one, which can be bit difficult to deal with- one that used a similar method for current sense by using the MOSFET for the current limit in the TPS40071/77 family.

    First, the accuracy of the internal reference in the chip from part to part, and the offsets of it's internal amplifiers- Notice how the TPS61260 datasheet is devoid of current accuracy information over temperature, and missing most of the important information an engineer should be considering - these sorts of omissions from datasheets usually indicate a huge red flag- an area that one needs to carefully consider and test carefully.

    At the currents you are showing, a flashlight would be a good application.  Realistically, the temperature of my flashlight can range from 7 degrees F and if it is warm out (say up in an attic on a hot summer day doing an inspection) it could be up to 120 degrees F ambient, and if operating for 15 minutes, internally it could reach 160 degrees F- easily.  This range could be much wider, depending on where in the world you are, and time of day.

    Since the part you mention is devoid of critical details, lets use the TPS40071:

    Internal Error amp feedback regulation voltage low of 0.690V and up to 0.715V

    Input bias current -250nA to 0nA (creates an effective error in the feedback divider)

    Skip to current limit offset voltage -75mV to -30mV over temperature.

    Add to that the current sink into the current limit, which varies from 80 to 125uA.

    Now, lets jump to an external MOSFET one may use, to give us an idea of what happens to the MOSFET, especially considering temperature. Lets use a common MOSFET datasheet, say the FDS6612A found here:

    www.fairchildsemi.com/.../FDS6612A.pdf

    In figure 3, you can see the temperature causes the on resistance to vary by a factor (1 being the value at 25C).  7 degrees F is -14 degrees C, and 160 degrees F is 71 degrees C.  However the die inside can be much hotter, and there isn't much room in a flashlight, and I've never seen the MOSFETs drain, or "thermal pad" attached to the body of the flashlight.  The boards are tiny, and they don't have room for even the minimum pad.  So lets say the die is cold at first, and eventually internally it heats up an additional 50 degrees C over 15 minutes.  So, we are looking at -14°C to 121°C (71 +50). So, using Fig. 3, we see the MOSFETs' on-resistance will vary from a factor of more than 0.9 to 1.4, about a 50% variation in on-resistance (depending on how you look at it).

    Now, on the same datasheet we jump back to page 2, and look at the Rds(on) which is specified at 25°C, lets use the 10V gate drive, nominal 0.019 ohms to 0.022 ohms max, with no mention of the minimum. From experience, the minimum can easily be the same amount lower, so lets use 0.016 ohms for that.  Remember, this is at 25°C, so we already get a 28% variation part to part, add to that the 50% percent on-resistance variation due to temperature, and you are looking at a 78% variation of the MOSFET on-resistance over temperature.

    Now, go back to all the errors accumulated in the chip's reference, bias current, and current limit offset and consider it all together.

    It gets even tougher if you need the current limit to always function as it should, and you need to specify in a MOSFET that can handle all the errors, without letting out the smoke- seriously increasing the cost of the design, and lowering the efficiency.

    It would be nice if the chip designers would consider these basic realities, and compensate for them.  However the designer probably figured the chip would end up in a cable/dish control box that sit's in a climate controlled living room, and assumes the designer has plenty enough room to put a 1" square patch of 2 oz. copper on the MOSFET, or has room a heatsink for the SO-8.  Although this isn't the case much of the time, were one may be limited by space, no flowing air, no heatsink, and has to use 1/2 oz copper due to a part that is only in a CSP scale package, or 0.3mm pitch BGA FPGA.

    Though the idea might work okay for a really low power flashlight that uses a T1-3/4 LED, but their price point is already so low that they use the button cell's own internal resistance to limit the current, since they can't afford the cost of all the additional electronics.  Though maybe it could make sense in a cell phone.

    Consider all this again when you have 350mA to 3.0 Amps flowing in the LED in common flashlights where the price point is high enough to have decent electronics.  Some of the high performance flashlights are driving 9 Amps into a Luminous Devices LED, or using an array of multiple high power LEDs from companies like CREE, Royal Philips LumiLEDs, OSRAM, etc...

    Just some thoughts to muse/chew on.

Children
No Data