Test & measurement
Multiphase buck converters are chosen to satisfy the power demands of the latest artificial intelligence (AI) processors. Checking power integrity presents extra challenges for engineers, over and above those associated with basic measurement, to properly check the performance during power supply validation and debugging.
A
I is transforming computing and substantially changing system power requirements. In addition to raising peak consumption, AI workloads
cause wide power fluctuations as computationally demanding floating-point and mixed-precision arithmetic happens in parallel on thousands of cores and involves intensive memory exchanges. Despite the changeable conditions, energy efficiency and stability are critical, everywhere from data centers and high-performance computing to AI accelerator cards and processors. Accordingly, the changes in power demands are pushing power supply design towards multiphase buck topologies, which are well suited to delivering high power efficiently while responding quickly to demand transients. Distributing the load across multiple phases reduces stress on individual components and reduces losses associated with high-current single-phase designs. In addition, spreading power conversion across multiple phases reduces heat buildup, improving efficiency and reliability. When a load step occurs, a multiphase controller can react quickly by changing the Pulse Width Modulation (PWM) signal for the next phase. It is also possible to
change the phase alignment dynamically and turn individual phases on or off as needed to adjust the power delivered across a wide range.
STATIC TESTING FOR RIPPLE AND NOISE Fundamental power integrity analysis is needed to ensure the power rail voltage complies with the required noise and ripple, using a variety of test loads to observe the performance across the range from minimum to full load. Analysis is needed in both the time and frequency domains (figure 1) to examine noise at the switching frequency and its harmonics. Using an oscilloscope that has a high acquisition rate makes it possible to see effects in the power supply with full clarity. The instrument used to capture the image in figure 1, the MXO 5 from Rohde & Schwarz, can capture up to 4.5 million waveforms per second.
For power supplies and converters of any type, measuring static performance is becoming a more exacting practice at the lower supply voltages required to power Integrated Circuits (ICs) built using the latest process geometries. The graphics processing units (GPU) or Tensor Processing Units (TPU) of today’s AI accelerator cards are operating from supply voltages of 1.0V
or less, sometimes as low as 0.6V. As the typical supply voltages are reducing, voltage tolerances are also tighter making power rail qualification a more challenging task. Even older standards like 5V and 3.3V rails may now have tighter tolerances than in previous generations. Given that a tolerance of 1% on a 1V supply means the rail could vary by up to 10mV, the probe needs to be able to measure in the single- digit millivolt range to capture fine details. This can be difficult with some types of oscilloscopes. In particular, a scope with limited offset can have difficulty zooming in on the waveform sufficiently. At these levels, noise from both the scope and the probing system will impact the measurement. It is not possible to measure signals that are smaller than the noise on the scope. Also, the oscilloscope and probe may not have enough bandwidth to see coupling sources and capturing enough waveforms to ensure seeing potential outliers can be difficult.
A high offset voltage setting in an oscilloscope or probe is needed to focus on small fluctuations in voltage while seemingly ignoring the larger DC component. By shifting the waveform vertically, adjusting the oscilloscope resolution can zoom in on the ripple, making small variations more
Figure 1. Power integrity analysis in time and frequency domains. Continued on page 30... Instrumentation Monthly January 2026 29
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73
