New Intela idea: next‑generation integrated capacitors will provide reliable power for future AI chips

Intel unveils a new generation of capacitors for the AI era chips

1. What changed – Transistors are always considered the key elements of a chip, but in the context of miniaturization and rising power consumption they require additional support from other components—primarily capacitors.

- Intel Foundry (the contract manufacturing division) announced the release of MIM capacitors (metal‑insulator‑metal), which promise a significant boost in power supply stability and capacitance.

2. Technical details | Metric | Current solution | New MIM solution

|---------------------|-----------------|-------------------|
| Capacitance density | ~20 fF/µm² | 60–98 fF/µm² (three‑fold or higher) |
| Leakage current | – | One order of magnitude below industry requirements |

- Capacitors are integrated into trench structures on the backside of the wafer, allowing use of existing manufacturing lines.
- A cylinder with a core inside (single dielectric layer) simplifies fabrication.

3. Materials‑manufacturers Intel identified three promising materials for MIM structures:

All three materials are ferroelectrics—they change their capacitance under an external field while maintaining parameter stability.

4. Practical value – Thermal stability: capacitors withstand up to 90 °C for 400,000 seconds (~110 h) without losing capacitance.
- Longevity: transistors are projected to operate without breakdown for 10 years even with constant over‑voltage—“a gift” for overclockers.
- Power efficiency: a significant increase in performance is expected for data‑center chips, mobile devices, and AI accelerators.

5. Conclusion – Intel Foundry aims to elevate power distribution technologies into the AI era by providing capacitors that not only scale with transistors but also deliver higher capacitance, lower leakage, and reliable operation under extreme conditions. This will accelerate AI chip development, improve energy efficiency, and ensure stable performance in the most demanding applications.