It’s time to look at the next wave: vertical power delivery and the relentless Analog Devices (ADI) innovation that makes it possible.
If you’ve joined us thus far, you know the challenge: AI wants more power, more often, in less space, with zero tolerance for error. Multiphase PoL improvements have taken us far, but what happens when even those innovations can’t keep pace with new generations of ultra-dense AI xPUs?
The Rise of Vertical Power: A New Paradigm for AI PCBs
Traditional power delivery runs laterally, with voltage regulators sitting somewhere off to the side, pushing current across precious PCB real estate. But with the need for 650 A continuous and over 1000 A peak delivery becoming standard, resistance across even a short trace isn’t negligible; it’s a showstopper.
Vertical power (sometimes called backside power delivery) upends the game. Instead of snaking currents through centimeters of PCB, the entire VRM (controllers, power stages, inductors, caps) is mounted on the backside of the PCB, directly underneath the xPU. How does it work? Power flows vertically, through dedicated vias, straight from the PoL VRM to the xPU’s power pins, cutting trace length, impedance, and voltage drop nearly to zero.
Figure 1: A Vertical Power Module Architecture (for illustration purposes only)
Benefits Beyond Density: Efficiency, Thermals, and Performance
Here’s why vertical power isn’t just about cramming more into less space:
- Reduced I²R Losses: Short, direct paths mean power reaches the xPU with less wasted energy and less heat, boosting both efficiency and reliability.
- Faster Transient Response: With less interconnect and tighter physical integration, voltage rails recover from AI’s wildest workload surges in a flash.
- Easier Thermal Management: High-frequency decoupling caps can sit right under the xPU, where they’re needed most, with heat flowing straight out of the device through top-side cooled packages.
The result? The holy grail: maximum power in minimum space, perfectly aligned with the AI revolution.
How ADI Makes Vertical Power a Reality and Why It Matters
Three core technologies converge to unlock vertical power’s full potential:
- High-Integration Flip Chip Packaging: Both the MAX16602 and MAX20790 are designed for minimal parasitic inductance and optimal thermal transfer, a must for backside VRMs.
- Patented Coupled Inductor Technology: ADI’s CLs maintain ultra-high efficiency and ultra-small size in vertical footprints, delivering both the current and transient response that today’s AI boards demand.
- Advanced Control and Monitoring: Pure hardware isn’t enough. Digital telemetry, autonomous phase-shedding, and fast-fault recovery keep tomorrow’s systems safe, predictable, and futureproof.
Eyes on the Horizon: What’s Next for AI Accelerator Power?
As AI architectures and applications continue their rocket-fueled evolution, ADI is already imagining what comes next. Even higher density VRMs are expected, scaling from 650 A continuous to over 1000 A supercards already emerging in bleeding-edge data centers. New generations of intelligent, reconfigurable control are being designed, including software-defined regulation, real-time telemetry, and AI-in-the-loop for self-healing power networks. Ecosystem partnerships will offer free coupled-inductor IP licensing, deep collaboration with leading magnetics vendors, and customer APIs for custom telemetry and analytics.
No matter where AI goes, ADI will be there: keeping power safe, fast, and precise with the smallest, coolest, and most reliable platforms on the planet.
Conclusion: Powering Tomorrow’s Intelligence Today
AI’s impact on society and the planet will only grow as new applications emerge in healthcare, climate, security, and beyond. The difference between a breakthrough and a breakdown will be the systems that make it work, starting with power delivery.
ADI is proud to be at the forefront, delivering performance, efficiency, and innovation at every layer: whether you’re building the next AI supercard, scaling cloud platforms, or inspiring students to light up the world’s hardest problems.
For more on this, see the technical article Impacts of Transients on AI Accelerator Card Power Delivery
Read all the blogs in the Powering AI Accelerators series.