VRM, not only DCx - 1000A/us
We recently measured very high efficiency and A/mm2 up to 1000A in the DPx, “split” converter.
https://lnkd.in/dYV6ejwt
Now we use the 500A version to measure its dynamic performance, and answer the question:
Is this converter capable to supply high di/dt load steps?
HPC & AI chips require very demanding load steps, e.g.:
➡ 100 A
➡ 1000 A/us
Our load setup reaches “only” 150A steps with “only” 150A/us.
This is much less than what the DPx converter can do, which is, theoretically, >1000A/us. We measured 650 A/us, limited by the slew rate of the current probe we used.
It is enough to measure what we want, though. The question is:
How can you test that your power converter is faster than your load?
Simple, rather than the typical “voltage drop”, which is so challenging to be limited to 60-70mV under a >100A load step, we can measure a “voltage rise”, if the converter is faster than the load.
The DPx converter injects a current step to the output capacitor, synchronized with the load step (you need a fast detection of the load step) and … voilà ...
Vout increases!
This means that Cout can be reduced, being mostly determined by voltage ripple rather than by the load steps !
💡 Imagine how much area is saved under the chip.
🔎 Not to mention cost savings.
What is next?
Improve the load test setup.
Can you imagine measuring a “voltage rise” under a 300A; >1000A/us load step in the 2000A “Emerald” converter?, switching at only 150kHz??
We do!
Teamwork: Alejandro Castro Molina, Pablo Mazariegos López, Alejandro Figueroa Pulido, Alvaro Cobos Figueroa, Javier Goicoechea Secilla, María Portela Díaz
#hpc #ai #power #powerdelivery #powerelectronics
Senior Analog Designer @Alphawave Semi | Ex-Synopsys | | Analog Circuits Design Engineer
3wCongratulations to the team...