Measuring current consumption can often be a bit tricky. In this lesson we will go through the most common pitfalls, different techniques we can use, and how to make sure that the data we get is accurate. We will learn how to prepare the hardware for current measurements, how to verify our measurement setup, and how to interpret the results.

The biggest problem when measuring current is that the results we are getting are simply wrong, but since we do not have any ways of verifying, lack of experience in identifying issues with the setup, or identifying obvious mistakes in the current profile, or consumption, we simply accept the result as it is and try to make sense out of it. We try to fit observations with our assumptions.

• First we need to learn about different measurement techniques, what we can expect from the results, what to use in different scenarios etc.
  • This is important so that we can identify weaknesses in the methodology.
• We need to have an idea of what is realistic in terms of current consumption. Ballpark numbers on active currents, like CPU and peripherals, vs idle current. uA vs mA.
  • This is important so that we can do an assessment of the validity of the results we get.
• Have a plan on what you want to measure, divide measurements into separate tasks. Which technique can be used for which part.
  • This is important so that the results we get have a value, that they can actually be used to improve the design, or identify real issues with the design

Dynamic range

No current measurement equipment is actually measuring the current. What it measures is the voltage drop over a series resistor, also called shunt resistor, or measuring shunt, which is placed in series with the load/DUT.

So in practice we have a resistor in series with the DUT, and we then measure the voltage drop over this resistor and convert it into a current using Ohms law, which states that voltage and current are proportional, with a given proportionality constant. And since the proportionality constant, in our case, is the value of the shunt resistor, which again is a predefined quantity, converting from a voltage drop to a current is a trivial task. I = V/R

The shunt resistor can either be added manually, in cases where we manually measure the voltage drop, for instance if we want to use an oscilloscope. Or the shunt is integrated into the measuring equipment, like ampere meters or power analyzers.

The next step is to choose the size of this shunt resistor. At a given current, a larger resistor will result in a larger voltage drop, making it easier for the measuring equipment to capture the correct current. But when the current increases, the voltage drop will eventually be too large. This means that our goal (or the measuring tool’s goal) is to choose a resistor value which will have minimal impact on the measurements, meaning as low voltage drop as possible, but still being able to measure it accurately.

The difference between the lowest measurable current, and the highest current which will not impact the DUT too much is called the dynamic range. So how large dynamic range do we actually need? On a micro controller we often have sub 1 uA sleep current, and run currents of 1 mA up to 100 mA. Meaning that we need a dynamic range that covers maybe 100 nA up to 100 mA. This is physically impossible, with a single fixed shunt resistor.

Let’s say that we have an application that has an active current of 10 mA and an idle current of 1uA. Since we have up to 10 mA current we can choose a 10 Ohm shunt resistor. A 10 mA current will give a 100 mV voltage drop over the shunt. V = IR. 100 mV will probably not impact the DUT too much. Choosing a 100 Ohm shunt on the other hand will result in a 1 V drop, which is too much. So our 10 Ohm seems like a good choice so far, and the largest resistor we can use in this case. But what happens when the current drops from 10 mA down to a 1 uA idle current, will we still be able to measure the current accurately? The voltage drop will now be 10 uV, which is too low to measure for any equipment. So in order to measure 1 uA you would probably need at least a 10 kOhm shunt (= 10 mV drop), in order to be able to measure it accurately. This will not add up, since we already established that 10 Ohm is the largest resistor we can use, because of our 10 mA active current.

So what do we do? We have two choices:

1. Use a tool that switches the measuring shunt automatically based on the current draw.
   • This is called automatic range switching, and is used in power analyzers/SMUs and some multimeters, often bench top multimeters.
2. Use a fixed shunt value, and select which current range area we will focus on. Either the idle current, or the active current.
   • This will be our method if we want to use an oscilloscope or multimeters which do not have automatic range switching. Often handheld multimeters where you select the range using a wheel or similar.

Issues with automatic range switching

If we choose to use a tool with automatic range switching we should be aware that the voltage source output impedance which change when switching the shunt, and it will impact measurements. How much depends on the tool. Some equipment is really good at compensating for this in software, others not so much. What happens is that when the shunt is switched there will be a sudden drop or increase in the voltage on the rail we are measuring. This will in turn lead to an inrush current going into or out from decoupling capacitors etc, giving an overshoot when there is a positive current transient, and undershoot when there is a negative current transient.

Sample rate

In order to capture the current consumption and convert it into a plot or a reading on a display, we need to sample the voltage over the shunt resistor. One sample is basically just one voltage measurement, which is either shown directly on the display, e.g. on multimeters. Or stored in a dataset, that can later be exported or plotted on a screen, e.g. power analyzers, oscilloscope.

The sample rate is how often we do this sampling. The higher the sample rate the more details we will get out of the current consumption plot, up to a certain limit.

In electronics circuits we often have many decoupling capacitors which are used as energy reservoirs, or to filter out unwanted noise. In terms of current measurements, these capacitors acts as filters, meaning that the time wise resolution will somewhat be limited. When increasing the sample rate, you will get to a point where increasing it any further will not add any details to the plot. So there will be a limit to how detailed the plots can be. Will show examples of this below

So how high sample rate do we need? It depends on our use case, so here is an example on how high sample rate you would need in order to extract different data points out of a Bluetooth LE advertising event.

	200 kSa/s	300 MS/s
Bluetooth LE advertising *200k samples per second is enough for extracting the TX/RX pairs from the event. *Due to capacitance on VDD the 300MS/s plot does not give us any more information

Using a handheld multimeter where you read the current on a display, the refresh rate on the display, and what is possible for the human eye to register will limit the sample rate to around 1 second.

Current and charge

When supplying a voltage to an active load, like an MCU, a current will flow which often will change over time. If we use a tool to plot the current over time, we then plot the instantaneous current, or in other words what the current is at a specific point in time.

We often talk about “current consumption” and a somewhat common misunderstanding is that we are now talking about a quantity, like saying that “my device has consumed 10mA”. But this is not correct. Current consumption is just the average instantaneous current over a given time, and the “quantity” in this case would be the average current multiplied by the time, which is the electric charge (Q, unit = Coulomb [C]). So if your device has drawn 10mA over the course of 1 hour, your device has “consumed” 36C (10mA x 3600s = 36C)

The unit for electric charge is Coulomb (C), but as we discussed above, charge is current (A) multiplied by time (s), meaning that we can also use As (ampere second) as the unit. If we then convert seconds to hours, and ampere to milliampere we get the more commonly used unit, mAh ( 1mAh = 3.6C), which might sound more familiar than coulomb. E.g. it is used by battery manufacturers to specify how much charge a fully charged battery can hold. In this lesson, however, we will use coulomb, C, instead of mAh.

So when do we use current consumption, and when do we use charge?

1. We use current consumption if we want to express the average run current of our device (running over an indefinite time). This is convenient because we can easily estimate battery life time. E.g. if our battery has a capacity of 100mAh, and our devices draws 10mA in average. The battery will last 10 hours (100mAh / 10mA = 10h)
2. We can measure the steady state current in different “operating modes”, like CPU run current, radio TX run current, idle current etc. We use this information to identify parts of our application that can or should be optimized. Or just to explain the current consumption in general, e.g. when evaluating different solutions.
3. We use charge when looking at sections of a plot, e.g. when looking at one Bluetooth LE event, or one CPU event running a specific algorithm or similar. We can then use this information to decide on the Bluetooth LE interval, or how often we can wake up to do this specific CPU task, while not using too much current in average. More on this in the Periodic signal and extrapolating data section where we use the charge to find the average current at different intervals

An important aspect here is that current consumption during an active event, like a Bluetooth LE advertising event, is not a very interesting parameter. It doesn’t really give us any information, because we also need to include the time frame. But if we include the time frame, as in multiplying with time, we get the charge.

Below are two plots illustrating this point. Both show the same Bluetooth LE event; the only difference is the placement of the cursors. In the left plot we get 2.0mA in average between the cursors, while in the right plot we get 1.5mA. Only because we include a bit more of the idle current within the cursors, the average current is significantly lower. This shows that recording the average current during a event like this has no value, unless we include the time between the cursors as well. But then again, current multiplied by time equals charge, so the information we really want here is in fact the charge, which you can see is the same in both plots. This is the information we use to calculate or estimate the total average current in the end.

Average current during event: 2.0mA Event total charge: 8.0 µC	Average current during event: 1.5mA Event total charge: 8.0 µC

Power and energy

Although not so common, we can also use power and energy instead of current and charge. They only differ by a factor, which is the voltage.

Here is a plot that shows a measurement of a Bluetooth LE advertising event with three TX packets, illustrating the relationship between the current and power.

• Yellow = Current
• Blue = Voltage
• Green = Power

So as you can see power is just the current multiplied by the voltage (in this case 3.0V), and energy is the charge multiplied by voltage.

So far so good. Things start to get a bit more interesting when we change the voltage.

Here we have the same Bluetooth LE event, with the three TX packets, running at 1.8V instead:

So what information can we extract from these two plots?

After adjusting down the voltage from 3.0V to 1.8V,

Inside the nRF54L SoC there is a DC/DC regulator that regulates the VDD voltage down to the internal operating voltage of the device (0.7V to 0.9V). If the regulator was an ideal regulator, we would have seen the same power/energy going into the device, regardless of the input voltage. And we would have seen that the current/charge increased with the same factor as the voltage decreased. So in the example above, the charge would have increased from 8.0µC to 8.0µC x 3V / 1.8V = 13.3 µC, if we had an ideal regulator.

It can be useful to think of power/energy being drawn inside the chip, instead of current/charge. E.g. sending a radio packet at a given TX power will always consume a given amount of energy. And all the energy needed to send this packet will have to come from the VDD supply. This means that the total energy in the plot above, the total energy going into VDD, is a sum of the energy used to send the TX packets, regardless of the VDD input voltage, and the efficiency loss inside the regulator.

And, as mentioned above, the current/charge will scale opposite to the voltage so that the same “amount” of power/energy is supplied on VDD. Since power equals current multiplied by voltage (P = I x V).

Then finally we need to take the efficiency loss inside the DC/DC regulator into account. At 3.0V we have a larger voltage drop over the regulator than at 1.8V, hence we have slightly higher energy consumption at 3.0V than at 1.8V, because of the efficiency loss. Which explains the 12% decrease in energy consumption when going from 3.0V to 1.8V in the table above.

So when designing our product, what is best? Do we want low current or low power? Since we often only talk about current consumption when evaluating and measuring different solutions, it would make sense that the current consumption is what we want to optimize. But an important aspect here is that when measuring current, we always choose the voltage first, so that it matches the target voltage of our final design. Or we just choose a common voltage (like 3.0V, as we do in the datasheet), and then convert the numbers later when we know the operating voltage of our design.

Ultimately, the total energy is what we want to optimize, so running at as low voltage as possible is best. But there are often other considerations to take. What is the operating voltage of other components on the board? Are they compatible with 1.8V? What is the battery voltage? If we use a 3.7V lithium battery we need an external regulator anyways, to regulate the voltage down to 1.8V, but will the efficiency of this regulator be as good as the internal regulator on the nRF54 when running at 3.0V? If we use a 3V coin cell battery, should we add another regulator stage to get the voltage down to 1.8V? If this regulator stage also introduces efficiency loss, we might be better of supplying the nRF54 directly from the battery at 3V. Often the supply voltage is decided by other factors in our design. But a good rule of thumb is that a lower voltage is more efficient.

Periodic signal and extrapolating data

When you have a periodic signal, instead of measuring the current over a longer time in order to obtain the average current, you can break down the measurement into smaller pieces. First you can measure the total charge of the event which occurs periodically, then measure the idle current, and use this to calculate the average current.

This can be convenient when you do not have access to a tool that can measure over a longer time period, or if you want to quickly get the average current at different periods, for example different Bluetooth LE advertising intervals, CPU processing intervals, etc.

Here is an example showing how we can use a oscilloscope to find the charge for a Bluetooth LE advertising event, and use a multimeter to measure the idle current. Then at the row below we can see that our method is giving us the correct results by comparing with measurements using a power analyzer.

	Event charge	Idle current	Average current
Oscilloscope and ampere meter			I_avg = Q_event / interval + I_idle
	Q_event = 6.8 uC	I_idle = 2.84 uA	I_avg = 6.8uC / 1s + 2.84 uA = 9.64 uA
Power analyzer
	Q_event = 6.8 uC	I_idle = 2.88 uA	I_avg = 9.69 uA

Using this information we can create a list of the average current resulting from different advertising intervals. This is useful when we want to evaluate different options, so we don’t have to do a lot of different average current measurements.

Finally, take note of how the cursors in the following plots are placed. The left plot will show the correct average current. While the right one is not correct. The measurement duration (in this case the distance between cursor 1 and 2) should be an integer multiple of the signal period. If you have multiple periodic signals with different periods, then the cursor distance should be an integer multiple of the least common multiple of both periods for accurate measurement.

Here the signal period is 100ms. And in the left plot we have a cursor distance of 300ms, which is a multiple of the 100ms period. Note that this number also matches our calculated number from the table above.

Correct	Wrong
avg: 71 uA	avg: 62 uA