Traditionally, power engineers are trained to think in the time-domain while RF engineers are trained to think in the frequency-domain. To successfully design high-frequency DC-DC converters, the engineer will need to jump back and forth freely between the two perspectives. High frequency design requires new attention placed on parasitic impedances, effects from PCB layout, and potential sources of large output ripple switching spikes, and radiated and conducted electrical noise. As switching frequency continue to increase in the drive to reduce power supply footprint, power engineers and system designers will be increasing faced with issues that are outside of their realm of expertise. This article will help shed some light on the high frequency design considerations when designing with the next generation of fast switch-mode DC-DC converters.
High frequency content, it's not just the fundamental
The trend towards smaller DC-DC converter footprint is driving the industry to ever higher switching frequencies. To understand the impact of high switching frequency on DC-DC converter design, it will be useful to examine the spectral content of the switching waveform. Using the principle of superposition, the switching waveform (Figure 1 (a)) can be decomposed into three unique waveforms: a somewhat ideal square-wave (Figure 1 (b)), a triangular wave that represents the non-ideal rising edge of the pulse (Figure 1 (c)), and a triangle wave that represents the falling edge of the switching waveform (Figure 1 (d)). The rectangular waveform is the primary contributor of energy at the fundamental of the switching frequency with the familiar sin(x)/x , or sinc(x) envelope. The fast rising and falling edges contribute energy at 1/τR Hz and 1/τF Hz.
Figure 1. Switching waveform decomposition and associated frequency spectrum
As an example, the Enpirion, EN5335QI 3A-Integrated Inductor DC-DC converter, operates at a nominal 5 MHz. A 5 MHz switcher will have a 200ns switching period. Good efficiency requires that the rise and fall times be very small relative to the overall switching period. For this particular part, the rise time is 2ns and the fall time is ~3ns. The 2ns rise time corresponds to a frequency contribution at 500 MHz and the 3nS fall-time corresponds to frequency content at 330 MHz. Figure 1 (e) shows the power spectral density.
By considering the switching waveform and its corresponding frequency components, we can better understand the nature of the noise that must be controlled.