# How Chips Are Using On-Chip Voltage Regulators to Save Power

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 26 Jun 2026, 10 min.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0074.mp3

In this episode of The Hardware Podcast, Lucas and Luna explore the quiet revolution in power delivery: moving voltage regulators from the motherboard directly onto the chip. They dissect why Intel's FIVR technology, first introduced in Haswell, is now being reimagined at advanced nodes for both CPUs and GPUs. The hosts explain how on-chip regulators reduce power loss, improve transient response, and enable finer-grained voltage islands—allowing designers to shut off unused logic blocks at the nanosecond scale. Key numbers include the efficiency gains (typically 5-15% total power savings) and the trade-offs in die area and thermal density. Lucas and Luna also discuss real-world implementations from AMD's Ryzen and Apple's M-series chips, and what this means for the future of heterogeneous integration and chiplets. A concrete look at a technology that touches every modern chip, yet remains invisible to most users. #OnChipVoltageRegulators #PowerDelivery #IntelFIVR #ChipDesign #VoltageIslands #EfficiencyGains #Semiconductors #HardwareEngineering #AMD #AppleSilicon #HeterogeneousIntegration #Chiplets #PowerManagement #AdvancedNodes #Technology #FexingoBusiness #BusinessPodcast #HardwarePodcast Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So there's this quiet revolution happening inside every chip you use, and it's not about more transistors or faster clocks. It's about how we get power to those transistors in the first place.
Luna: You're talking about voltage regulators. Specifically, moving them from the motherboard onto the chip itself.
Lucas: Exactly. On-chip voltage regulators. And the reason this matters is straightforward: every time you move power across a motherboard trace, through a socket pin, and then across the package, you lose energy as heat. The further power travels, the more you lose. So if you can generate the right voltage right where it's needed, on the die, you save real power.
Luna: And you also get faster response when workloads change. A motherboard regulator might take microseconds to ramp voltage. On-chip, you're talking nanoseconds.
Lucas: Right. That speed opens up something called voltage islands—different parts of the chip running at different voltages, even shut off completely, all independently and almost instantly. It's a huge lever for power efficiency, especially in mobile and data center chips.
Luna: Before we dive deeper—if today's tech conversation gave you something usable, something that sharpens your sense of where hardware is headed, a handful of listeners already chip in monthly through buy me a coffee dot com slash fexingo. That's what keeps this show ad-free and lets us spend time on topics like this.
Lucas: And we really mean it when we say a handful. It's a small group, but it makes a real difference. So if that feels right to you, we'd appreciate it.
Luna: Alright, back to those voltage islands. Lucas, you mentioned Intel's FIVR technology—Fully Integrated Voltage Regulator. They introduced that back in 2013 with Haswell. Why is it only now becoming a mainstream design pattern?
Lucas: Great question. The short answer is that the trade-offs were too steep at the time. FIVR added significant die area—Intel dedicated a decent chunk of the chip just to the regulators. And that area had to be integrated into the same process as the logic, which wasn't optimized for the analog components needed. So the efficiency gains were there, but they came at a cost.
Luna: And that cost included thermal density. You're essentially putting power conversion—which generates heat—right next to the hottest parts of the core.
Luna: So it's not one-size-fits-all. Which companies are actually using on-chip regulators today in shipping products?
Lucas: AMD has been a big proponent. Their Ryzen chips, starting with Zen 2, integrated voltage regulators on the I/O die, which is built on a more mature process. That allowed them to fine-tune power delivery to the chiplets. Apple's M-series chips also use on-chip regulators extensively—they have multiple voltage domains across the CPU, GPU, Neural Engine, and memory controller.
Luna: And I believe even some mobile SoCs from Qualcomm and MediaTek are using them, though they're less public about it.
Lucas: Right. The trend is accelerating because of two things: the move to chiplets, and the need for extreme power efficiency in AI accelerators. When you have multiple dies in a package, each with different voltage requirements, relying on a single motherboard regulator becomes impossible. You need local regulation.
Luna: Let's talk about the actual efficiency numbers. What kind of power savings are we seeing?
Lucas: Industry papers from Intel and AMD quote total system power savings of five to fifteen percent, depending on workload. That might not sound huge, but in a data center where a single rack can draw twenty kilowatts, a ten percent saving is enormous. For a smartphone, it translates to an extra hour of battery life.
Luna: And those savings come from two main sources: reduced I-R drop—the voltage drop across the power delivery network—and the ability to run each voltage island at its optimal point.
Lucas: Exactly. The I-R drop is the big one. On a large chip, the voltage at the center of the die can be significantly lower than at the edge because of the resistance of the metal layers. By having regulators distributed across the chip, you can compensate locally. That means you can run the global supply slightly lower, saving power across the entire chip.
Luna: What about the analog design challenges? These regulators are switching converters, right? They generate noise.
Lucas: They do. Most on-chip regulators use switched-capacitor or low-dropout linear regulators. Switched-capacitor ones are more efficient but introduce switching noise that can couple into sensitive analog circuits. Designers have to be very careful about layout—isolating the regulator areas with guard rings and dedicated power grids.
Luna: And LDOs are simpler but less efficient, especially when the input to output voltage difference is large.
Lucas: Right. So a common approach is a hybrid: use a switched-capacitor regulator for coarse conversion, then a small LDO to clean up the output. That gives you efficiency plus low noise. IBM demonstrated a design like that a few years ago, achieving over ninety percent efficiency.
Luna: Let's zoom out. How does this tie into the broader trend of heterogeneous integration?
Lucas: Heterogeneous integration is all about mixing different dies—logic, memory, analog—on a single package. Each die might have its own optimal voltage. With on-chip regulators, you can supply a single external voltage—say one point eight volts—and let each die generate its own internal voltages. That simplifies the package and motherboard design enormously.
Luna: And it enables tighter integration. For example, you could put a power management die right next to a compute die, with micro-bumps connecting them.
Lucas: Exactly. Some companies, like TI and Analog Devices, are selling dedicated power management dies that are meant to be co-packaged with processors. Basically, a tiny voltage regulator die that sits alongside the main chip.
Luna: So we're moving from a world where the voltage regulator is a separate component on the motherboard, to one where it's integrated into the package, and ultimately onto the die itself.
Lucas: That's the trajectory. And it's not just about saving power—it's about enabling new architectures. For example, near-threshold computing, where logic runs at very low voltages to save energy, requires extremely tight voltage regulation. Without on-chip regulators, it's nearly impossible.
Luna: And with the end of Dennard scaling, power efficiency has become the primary design constraint. So this technology is only going to become more important.
Lucas: Absolutely. We're seeing research into integrated magnetic inductors on silicon, which would allow more efficient dc dc conversion at high frequencies. There's also work on using capacitors in the back end of line layers to store energy locally.
Luna: I've seen some papers proposing on-chip supercapacitors made from advanced dielectrics. They could provide burst power for AI accelerators.
Lucas: That's the frontier. The dream is a chip that has its own power supply built in, almost like a tiny grid. But for now, the real-world impact is already significant. Every modern laptop, phone, and server chip uses some form of on-chip regulation.
Luna: Let's bring it back to something concrete. If someone is designing a chip today, when should they consider integrating voltage regulators versus keeping them off-chip?
Lucas: It depends on the power budget and the number of voltage domains. If you have more than three or four distinct voltage requirements, on-chip starts to make sense. Also, if you're using chiplets, it's almost mandatory. And if your workload has rapid power transients—like a GPU switching from idle to full load—on-chip regulators can respond much faster, avoiding voltage droop that could cause errors.
Luna: What about cost? Adding regulators adds die area, which increases cost.
Lucas: It does. But the cost can be offset by savings in the package and motherboard. You might eliminate several external voltage regulator modules and the associated passive components. For high-volume products, the net cost can be neutral or even lower.
Luna: So it's a system-level trade-off. Not just a chip-level decision.
Lucas: Exactly. And that's why you see it more in integrated products like Apple's M-series, where they control the entire system, than in commodity PC components. Apple can optimize across chip, package, and board simultaneously.
Luna: One more thing—how does this affect reliability? Adding more active devices to the chip means more potential failure points.
Lucas: That's a valid concern. But the regulators are typically built using the same proven transistors as the rest of the chip. And because they're operating at lower voltages and currents than external regulators, they can be quite robust. Companies like Intel have done extensive reliability testing on FIVR and found no significant impact.
Luna: I think the key takeaway is that power delivery is no longer an afterthought. It's becoming a first-class design consideration, just as important as the logic itself.
Lucas: Exactly. And the next time you look at a chip die shot, look for the small, regular structures near the corners or along the edges. Those are likely voltage regulators. They're small, but they're doing a huge amount of work.
Luna: And they're one of the reasons your phone lasts all day and your laptop doesn't melt your lap.
Lucas: Right. So that's the story of on-chip voltage regulators. A boring-sounding component, but it's quietly reshaping how every chip is designed.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.