In one of my previous posts, I went into the details of power integrity and the reason why we need to take it into account in both current and future hardware designs. Furthermore, I promised to provide some tips and hints on how to do this measurement correctly to make sure that the obtained values are actually from your design and not from the scope and the probe.
Before I go into the details of assessing power integrity correctly, I would like to repeat briefly why we have to care about this kind of measurement. Power rails in today’s hardware designs are affected by periodic and random deviations (PARD) caused by noise, load responses and coupling from adjacent traces like clock and data lines. All these deviations have the potential to violate the noise margins of individual components within the hardware design. Deviations can also corrupt data transmissions, because they propagate into the data lines. More details on PARD, its causes and consequences can be found in this post.
10 MHz clock signal coupling to the power rail on a hardware design.
So, what is needed to make a power integrity measurement as accurate as possible? Besides an oscilloscope and some kind of probing solution, of course. There are basically four requirements for the measurement solution: low noise, low loading for the DUT, DC-offset compensation and sufficient bandwidth. As one can imagine, these requirements are connected with each other and, to some extent, depend on each other.
The reason why low noise is important for power integrity assessments is straight forward. If a chip has a 3 % noise tolerance band around its specified supply voltage level, the signal we are interested in is in the range of a few tens of millivolts. A lot of noise from the measurement solution itself would just cover our signal of interest.
Obviously, the used oscilloscope and probing solutions are the main cause of noise that we are not interested in. So the requirement is simple: no 10:1 probing solution and the usage of the 50 Ω path on the scope. Why?
10:1 probes add a lot of noise due to their large internal resistance of 10 MΩ together with the scope. In addition, 10:1 probes are very sensitive to radiated emission.
A 10:1 probe leads to significantly more noise than a 1:1 probe.
The second need is the usage of the 50 Ω path on the scope. Thermal noise on the 50 Ω path is a lot smaller than on the 1 MΩ path, simply due to the fact that thermal noise depends on the resistance linearly.
In brief, use the 50 Ω path of the scope and avoid 10:1 probes for power integrity measurements to minimize the noise generated by the measurement equipment.
What do I mean by low loading? Commonly, this term is associated with the capacitive and resistive loading of a DUT when you probe it. Here however, the term has to be interpreted a bit differently. The goal of the probing solution should be to avoid a DC current flowing into the scope.
As you can see, this requirement conflicts with the low noise requirement and the usage of the 50 Ω path to some extent. It is quite obvious, that the 50 Ω path would draw a substantial amount of current from the DUT. As an example, a 3.3 V power line would lead to a current of roughly 66 mA into your scope. This current would not destroy your scope, but it can reduce the operating voltage of your DUT, it could discharge the battery of the DUT quite fast or just put your low dropout regulator into trouble.
As a smart user you would probably suggest a DC block to avoid DC currents flowing into the scope. But in the next section you will see why we should stay away from using a DC block.
To sum up the low loading requirement, our probing solution should block DC currents flowing into the scope even when we use the 50 Ω path.
One particularity when probing a power rail is that you would like to analyze small AC components which ride on a large DC voltage. Often the AC components are 50 to 100 times smaller than the DC part. This requires us to set the vertical resolution of the scope to only a few millivolts per division, which in turn prevents us from using large DC offsets. Often the limitation already kicks in after you moved the ground line a few hundred millivolts out of the screen. Ideally we would like to have a probing solution which allows the setting of a DC offset of up to 24 V and at the same time is able to use millivolts per division.
A commonly taken approach is the usage of a DC block. But I’d advise you against doing so due to two reasons.
The first reason is that a DC block not only blocks the DC components, but also the frequency components in the sub-hertz region. But in many cases these sub-hertz parts are of strong interest. For example, changes in the supply voltage caused by thermal heating of components on the DUT happen comparably slow and would be covered by the DC block.
A DC block removes sub-hertz frequency parts and covers slow changes of DC voltage on a power rail.
The second reason is not related to a measuring issue, but rather to damage prevention. Adapting or removing a DC block from a scope poses a certain risk. Charging or discharging currents have the potential to destroy the analog front-end of an oscilloscope. And believe me repairing this kind of damage is costly.
In summary, our requirement for the probe is to enable large DC offsets (ideally up to 24 V) while being able to use small millivolts per division settings without the usage of a DC block.
The demand for sufficient bandwidth arises from todays and future hardware designs. Clock and data frequencies in the ranges of several hundred megahertz cause disturbances in these frequency ranges. Therefore, it is important that the probing solution is able to capture these signal parts and hence requires sufficient bandwidth preferably in the gigahertz range.
Only with sufficient bandwidth, you can see all spikes riding on your power rail.
How to probe?
We have seen that there are valid reasons why we should have a probing solution that has low noise, a low loading for the DUT, sufficient offset compensation and bandwidth. With standard scope probes such a solution is not available.
However, the Keysight N7020A power rail probe incorporates all these requirements. It is a 1:1 probe, uses the 50 Ω path, blocks DC currents but still senses the DC offset voltage up to +/-24 V and provides 2 GHz of bandwidth.
My first advice is that you check the power integrity on your DUT during development because it is one of the many sources for hardware issues which are not easily uncovered. To convince yourself, take one of your DC voltage regulators you commonly employ and take a look at its output with an oscilloscope. Even if you do not use an ideal probing solution, it is very likely that you will see noise and some signals caused by internal switching of regulating transistors.
My second advice is that if you want to check the integrity of your power rail, use the right tool. Otherwise you will most certainly not find the phenomenon you are looking for because it is blocked by your DC block, covered in noise or damped due to insufficient bandwidth.
If you would like to see this probe and its advantages in action, do not hesitate to give us a call or drop an email.