Time for another installment of the Easy Electrons series. This one is about the why’s and how’s of pull-up and pull-down resistors.

There are many ways to generate digital output signals. They almost always work with “high” and “low” voltages – for suitable definitions of high and low. As it turns out, there are lots of tricks you can play with this.

The high-end solution is to use a “push-pull” driver with two transistors. Similar to the H-bridge circuit described in an earlier post. By switching either one of the transistors on, you get an output which can carry current in either high or low states (“sourced” to ground, or “drained” from the supply rail, respectively):

The CTRL A & B pins are set up in such a way that one transistor conducts, and the other blocks. When a PNP/NPN pair is used, tying them both together turns out to work exactly as needed. This is what an Atmega pin does in OUTPUT mode, essentially.

But you don’t really need such a setup in many cases. That upper transistor can also be replaced by a resistor:

Resistors are cheaper, but more importantly, resistors offer more flexibility. The effect is that when the only remaining lower transistor is conducting, then it draws the signal to ground, and when it is not, then the resistor pulls the now-floating signal to a high level. This works as long as you don’t draw too much current. A common value for such a “pull-up” resistor will be between 1 kΩ and 100 kΩ. In that last case, you can’t put a load of more than a few microamps on the circuit, but again: for signaling purposes that is often just fine.

Why is this useful?

Well, for one, the voltage levels can be different. It’s possible to control a 0..12V digital signal using just one transistor and a pull-up to… 12V. This is also perfect for 3.3V -> 5V conversion, for example, and it even works when the output voltage is lower than the signal input (which goes through a current-limiting resistor anyway).

Another nice property is that you can tie several such outputs together. The output signal will drop to 0 whenever any output is low, and stay at 1 when all the outputs are high. i.e. it acts as an implicit AND-gate. This makes such a circuit suitable for “bussing”, i.e. putting multiple devices on a single signal line. You still need to collaborate between the devices to make sure only one of them talks at any given time, but electrically they can all be tied together with no extra circuitry. An output which is high is in effect not participating in the state of the bus signal.

Third, the single-transistor can still drive larger currents, but only in the “0” state. So you can still use this to turn on a LED (+ resistor), as long as you tie the LED/resistor combo between output pin and the supply voltage, not between output pin and ground.

And fourth, perhaps not as important as the rest: shorting such an output to ground when it only has an active-low transistor cannot harm the circuit (unlike shorting it to the positive supply rail). All that happens is that the pull-up will be supplying a little bit of current as it ends up getting the full supply voltage.

Note that putting a positive voltage on the transistor base willl cause it to conduct, and than tying something between the output pin and the supply voltage will cause it to get power when the transistor is conducting current from the output pin to ground. So loosely speaking, a “high” input voltage leads to an energized output state, even though the effect is to pull the collector low.

A slight variation is to use one transistor and leave the resitor off altogether. This is called an “open-collector” (OC) output, for obvious reasons. Sometimes, that extra resistor is not needed, i.e. when all you care about is energizing a lamp / motor / LED, or not. Sometimes a pull-up is still required, but then a single one will be sufficient, even with multiple OC outputs tied together. This is a way to reduce the total component count a bit further.

The I2C bus is an example which uses the open-collector output plus pull-up resistor approach, to implement a bus with multiple devices, exchanging information between them in any direction.

On an ATmega, you can enable a pull-up resistor on any pin by setting the pin up as an input pin, while writing a “1” as if it were still an output pin. The result will activate a weak pull-up resistor of 20..50 kΩ. One side-effect is that the input pin will read as a clear “1” when nothing else is connected. A pull-up resistor is a bit like slanting the table, i.e. the signal tends to stay in the high state unless a certain amount of current pulls it low.

What about pull-downs?

Technically, all this pull-up trickery can also be done with pull-down resistors and transistors tied between supply and output pin, but it is less comoon to do so. Due to the way transistors work, such a “high-side” switch is easiest to implement with a PNP transistor – and these used to be less common and more expensive. Furthermore, a “high” input signal would now cause no output current to flow. So the sense of operations is inverted. Are these essential reasons to avoid such an approach? No. But the NPN low-side switch approach has become prevalent. It has become second nature to use NPN transistors, and to tie small loads directly between the output pin and the positive supply voltage.

The ATmega has no built-in pull-down mechanism.

Yesterday was about MOSFETs and the heat they generate. Pretty impressive, those components which can control a 15A current with just a little bit of voltage.

Unfortunately, it’s not quite as simple as that. We have to check what happens with a lowly 3.3V applied to the gate. This can be found in the IRLZ34N datasheet:

This is an important graph for MOSFETs. Each line corresponds to a different voltage applied to the gate – this graph is in fact eight graphs in one.

You can see that they really do have some limits at lower voltages. At 3.3V, the IRLZ34N will not go much higher than 7A – quite a bit lower than the 30A maximum specs on the front page of the datasheet. Most MOSFETs are still at the end of their range when driven by a 3.3V microcontroller.

What this graph also shows, is the drain-to-source voltage at different current levels. This makes it very easy to estimate power consumption: at 3V and roughly 5.5A, that voltage will be 1V. In other words: 5.5W – quite manageable if the MOSFET is mounted on a suitable heat sink. But again: quite a bit lower current handling capacity than the 15A from yesterday. Note that we could drive it at a higher voltage by adding an extra stage in front, using either a BJT transistor or a MOSFET to get better current handling capacity.

As you can see, there is a very distinct switch-over point at each gate voltage level, where the MOSFET stops conducting more current. Under the switch-over point MOSFETs act essentially like a pure (low-value) resitor, but above that point they start acting more like a current limiter. This has major implications for power consumption, since the drain-to-source voltage will rise.

With this particular IRLZ34N, driving it at 3.3V, it is best not to push it beyond about 5..6A, and to use a heat sink when power consumption rises above say 1W (i.e. at 62°C/W, that makes the bare MOSFET rise 62° above ambient). Looking at the 3V line in the graph, I’d estimate that this MOSFET would work just fine without heat sink up to at least 2A, perhaps even 3A.

But what if we want to regulate the current? I.e. what if we want to use the MOSFET in an analog manner, and not just as on-off switch?

We could lower the gate voltage somehow, to force the current down. It won’t be a linear relationship, and I’m not even sure a MOSFET will conduct any current at all when the gate voltage is lower than 1 or 2V.

But in general, it’s a bad idea. The reason is (again!) power consumption. Say we use the following circuit:

(the resistor is a pull-down, explained in a future post)

That’s a 12W incandescent light bulb, powered by 12V, and controlled by a MOSFET, in other words: it draws 1A when full on. Let’s say we want to dim the light by halving the current through it. The lamp acts like a resistor (it just get pushed so hot that it starts glowing, that’s all). So half the current is what you get when you apply half the voltage to it. In this case 6V.

Suppose we figured out exactly what gate voltage to apply to get a drain-to-source voltage over the MOSFET of 6V. That would essentially accomplish our goal: the lamp would be dimmed. Another way to look at this, is that we’re tweaking the gate voltage until the MOSFET acts like a 12 Ω resistor between drain and source. Then the lamp and the MOSFET both get half the voltage.

What about power? Well, the MOSFET will have to have 6V across it, as we just saw. At 0.5A, this means that it will have to burn 6 (V) x 0.5 (A) = 3 Watts of power.

This is totally counter-intuitive: when we switch a lamp full on, the MOSFET will consume less than 0.1 W (as gleaned from the graph), but to dim it, that same MOSFET would need a heat sink!

There is some logic in this, though, if you think about it. The 12V supply voltage is a given. So if we want to apply less voltage to the lamp, we have no other choice but to waste the excess energy. Which is exactly what the MOSFET (or BJT transistor for that matter) will do. So although we’re reducing the total power consumption in this circuit by halving the current, we’re forced to do so by wasting power – as non-visible light, i.e. heat.

Can we do better? Yes, fortunately, we can.

This is where “pulse-width modulation” (PWM) comes in. Instead of eating up the excess power, we can take advantage of the fact that incandescent lights are fairly sluggish in their response. What we do is pulse the power on and of in very rapid succession. So the lamp will constantly heat up and cool down, and the result is that it won’t be burning at full brightness.

Why is PWM so incredibly useful? Several reasons:

• we don’t need a way to generate a regulated analog gate voltage, we can simply generate digital on-off pulses with no extra circuitry needed
• the MOSFET is again being used as pure on-off switch, and remains maximally efficient – so we probably won’t need a heat sink
• power is no longer wasted, it is now effectively throttled instead – in very short and rapid bursts

It turns out that PWM works in a lot more cases than just incandescent light bulbs. DC motors are also sluggish, so controlling their speed with PWM also works extremely well. And better still, even LEDs work well with PWM, even though they respond instantly – because it turns out that our own vision is sluggish too!

See also an earlier post about PWM on this daily weblog.

So there you go. MOSFETs are the workhorses of power control, due to their incredible properties, and PWM is the technique of choice when it comes to throttling power devices (lights, heaters, motors, and more).

To continue this Easy Electrons series, this time I will go a little bit into MOSFETs.

To me, MOSFETs rank high up there, next to operational amplifiers as one of the foundation components in electronic circuits which are extremely useful and practical. The most amazing detail is that MOSFETs only exist since a few decades – newbies when you consider the time scale of most electrical components.

In a way, a MOSFET is like a BJT transistor. Even the symbol for it looks similar (from Wikipedia):

This “N-channel enhanced MOSFET” type is the most common one, and like the NPN transistor, current flows into the top (D = drain) and comes out the bottom (S = source), all controlled by a third pin (G = gate).

One key difference between a BJT transistor and a MOSFET, is that a BJT is driven by current, whereas a MOSFET is driven by voltage. You feed a (small) current into a transistor to make it conduct, whereas you apply a (small) voltage potential to make a MOSFET conduct. The gate of the MOSFET doesn’t conduct current – it just senses the voltage. Bit like a capacitor.

You could say that a transistor is like a water wheel: you have to keep churning the crank to keep the water flowing through it. Whereas a MOSFET is more like a flexible tube: you pinch (well, “un-pinch”) to control the flow, but that pinch doesn’t consume energy. You could use a small mechanical clamp to maintain the pressure and keep the flow going.

This also explains why a transistor won’t conduct if the base is left unconnected (no current coming in), whereas a MOSFET could be doing anything when its gate is left unconnected, depending on how much charge was left when last connected. Early MOSFETs were in fact incredibly sensitive to static electricity – just touching the gate with a finger would often destroy a MOSFET. Nowadays, they are ESD protected.

MOSFETs are perfect for controlling large currents via a microcontroller. Even the weakest output pins can drive them, as long as the voltage is high enough. In the past, MOSFET’s needed at leat 4.5 to 5V to make them conduct, but nowadays voltages in the 2.5..3V range are sufficient in these so-called “logic-level” MOSFETs.

I’ll take the MOSFET Plug as example. It has two MOSFETs tied directly to two ATmega output pins:

Let’s look at a manufacturer’s datasheet for that “IRLZ34N” MOSFET, because there’s a lot of useful information in there.

The IRLZ34N datasheet is a great example. Seven pages, full of details, facts, graphs, circuits, pinouts, drawings, etc. It’s worth getting used to reading datasheets. They are loaded with info. To me, datasheets are the user interface of electronics. Give me a part number, and I’ll grab the datasheet to understand what it can do.

Here’s the first part:

• logic level – aha! it can probably work with 3.3V
• V_DSS – that must be the max switching voltage, 55V .. plenty!
• R_DS(on) – will come to that in a minute
• I_D – max current through the drain, a whopping 30 amps
• 175°C – looks like it can witstand scorching hot temperatures

Ok, there’s your helicopter view of this component. What I’m interested in is: will it be able to switch my <insert-some-high-power-device-here> ?

Well, we’ve seen the max voltage and current specs. But what really matters is power consumption. Because that’s the heat that gets generated, and that’s usually what breaks things.

Power is voltage x current (E x I). The voltage in this case is the voltage across the MOSFET. But we don’t know that – not directly. What we do know is its “RDS(on)” – this is the resistance between drain and source when turned on. Heh, how obvious. And exactly the value we want. It’s a mere 0.035 Ω.

Ohm’s law says V = I x R, so the voltage across the device is the current through it times its resitance.

Combine these two and we get P = E x I = (I x R) x I = I x I x R. Power consumption is proportional to the square of the current. Aha – that explains why large currents can be so destructive!

Let’s try this. Let’s go all out and push 30 amps of current through the MOSFET. Its power consumption will be 30 x 30 x 0.035 = 31.5 Watt. That’s a fair bit of heat (small lightbulb).

But will it work?

To find out, we need to do a thermal calculation. What’s going to happen to those 31.5 Watt of power? Well, they will come out as heat, but how much heat?

Time to look at another bit of info on the datasheet:

Let’s take the last value first: R(theta)JA, or Junction-to-Ambient thermal resistance = 62°C/W. In other words, each watt of power at the junction (i.e. inside the MOSFET package) will leed to a whopping 62°C temperature rise when the component is suspended in “ambient”, i.e. free, air.

Hmmm: 31.5 x 62 = nearly 2000°C. Yikes, our MOSFET is going to evaporate!

What we need to do is mount this part on a massive heat sink to make sure those temperatures are never reached, by drawing the heat away and keeping the MOSFET (relatively) cool.

Fortunately, there are two other values. The way these work, is that they tell you how much “heat resistance” there is when it flows away from the junction where all the heat is being generated. And it’s really easy to work with:

1. draw a picture of the MOSFET and how it’s mounted
2. find out the heat resistance in each step
3. add them up to get a combined °C/W value
4. re-calculate 31.5 x <that-value>
5. make sure it stays under the max temp you want stay under (175°C would be too hot for a plastic case, for example – or even for a printed circuit board)

I’ll use a quick example, just to see how far we can push our MOSFET. Let’s assume the MOSFET is mounted on a (very good) heat sink which has only 5°C/W. Then we add up: 2.2 (junction to case) + 0.5 (case to heat sink) + 5 (heat sink to air) = 8.7°C/w.

With 30A current, we get 30 x 8.7 = 261°C. Whoops, can’t be sustained without damage.

Ok, let’s aim a bit lower: 15 amps. Now the power consumption becomes: 15 x 15 x 0.035 = 7.9 Watt. Without heat sink: 7.9 x 62 = 490°C – still way too hot, but with heat sink we get 7.9 x 8.7 = 69°C.

This value is a relative value. It means 69°C above the ambient temparature. So in a 25°C room, the whole thing would become 94°C. Still very hot, but not a problem for the MOSFET!

In other words: take a MOSFET, mount it on a big heat sink, and you can see how a tiny little microcontroller could control a 15A light or a motor which has a 15A peak current. That’s what makes MOSFETS so magical…

Careful with heat sinks, though. To get it right, you really have to include all the paths to “ambient”. If you mount the heat sink in a big box (which can withstand 94°C), then the temperature inside will rise. And those 69°C we calculated will make the whole setup rise accordingly! – it doesn’t take much to get a “thermal runaway”: core heats up, ambient heats up, core heats up further, etc. Until disaster strikes. Not quite Tchernobyl, but hey… be careful.

Soooo… what started out as a MOSFET introduction, has turned into a power and heat calculation. As you can see, it’s not very complex. It’s not a bad idea to find out up front whether a power circuit will self-destruct or not. Now just add a 2x safety margin, and you should be OK. Or better still: build the circuit, and confirm that the results match these predictions, especially under stress and near design limits.

Ehm… I’ve swept a little detail under the rug:

These calculations assume that we’re driving the gate to 10V, but we’ll only be applying a feeble 3.3V or so. Whoopsy daisy. Let’s go into that tomorrow.

With SMD becoming more and more common, I really wanted to get a supply of different resistor values – a bit like this binder with through-hole components:

So I decided to get 100 of each of the E12 series, thus named because it has 12 values per order of magnitude – spread out in a logarithmic scale: 10, 12, 15, 18, 22, 27, 33, 39, 47, 56, 68, 82.

Here are all the values from 1 Ω to 10 MΩ:

Nothing fancy, but pretty low cost. And as a bonus: 0 Ω resistors! Don’t laugh, these can actually be quite handy as bridges over other traces. You could even use a single-sided PCB by adding a couple of these as “wire jumpers”!

Here’s the final result, as stored in the lab. I keep a set of 100 Ω, 1 kΩ, 10 kΩ, and 100 kΩ around for quick-and dirty uses, but this way all the usual values are available when I need ’em:

Each drawer has 3 compartments, with two values of the E12 series in each. So that’s two drawers per order-of-magnitude (“decade”), times 7, plus the extra 0 Ω and 10 MΩ. All that’s left, is to add a couple of labels on the drawers to quickly pick the right one.

Ready to do lots more experiments here at JeeLabs! :)