Will Generative AI Drastically Change the World? -The Evolution Required for AI Services and the Capacitors Supporting Them - (Second Part)

Role of Capacitors in Processors

Capacitors are used in a variety of areas regarding hardware for generative AI and those used in the power lines of processors that perform key computational processes play a particularly important role. Just as a human needs food to move and think, a processor needs a sufficient supply of current to perform computational processing. The "power supply" is responsible for supplying this current. The current is carried through the power supply line to the processor, which often requires such high speeds and large currents that the power supply speed cannot keep up. If the supply fails to keep up, the voltage required for stable processor operation will drop significantly and operation of the processor will stop. As a result, the computational processes for learning and inference will also stop. In order to prevent such accidents, "capacitors" supply high-speed current to the processor. However, even when capacitors are installed, if their capacitance is insufficient, the voltage will drop for the same reason and the processor will stop operating (Figure 1-A).
Many of the processors used in generative AI require high current, and there is a concern that such conditions may actually occur. Thus, using large-capacitance capacitors such as Panasonic Industry's conductive polymer capacitors to supply sufficient current prevents voltage drops and enables the processors for the generated AI to continue long-term and stable computational processing. (Figure 1-B)

A:When conductive polymer capacitors are not used
  → Current supply is insufficient and the voltage drops.
  → The voltage cannot be stabilized and it deviates from the acceptable range.
  → The system shuts down and data processing is aborted.

B:When a conductive polymer capacitor is used
  → Sufficient current is supplied, preventing voltage drop.
  → The voltage remains at an acceptable range. = The voltage stabilizes.
  → Operation can smoothly transit to higher-speed processing.

Features of Conductive Polymer Capacitors and Their Contribution to Processors for Generative AI

<Comparison with other types of capacitors>
In power supply design, combining multiple capacitor types and ratings according to design requirements is considered as the optimal solution. What are the advantages of conductive polymer capacitors over other types of capacitors when they are used for processors for generative AI? Conductive polymer capacitors are recognized as very well-balanced capacitors, offering the general advantages of low ESR and high safety for general electrolytic and tantalum capacitors (Figure 2 left) as well as stable capacitance and a fewer number of components for ceramic capacitors (Figure 2 right). Due to these benefits, conductive capacitors are widely used in power supplies for processors for generative AI that require stringent electrical characteristics and reliability.

General electrolytic and tantalum capacitors	Ceramic capacitor (MLCC)
One-digit lower ESR and impedance ➡Suppress voltage fluctuation at high-speed current Excellent safety ➡Contribute to highly-reliable design	Stable high capacitance that is not influenced by voltage or temperature ➡Prevent voltage drop caused by insufficient power supply A fewer number of components ➡Contribute to simpler design

Compact and Reduced Height

There is a tendency for AI processors to become larger each year as large currents are required to perform high computing capabilities. The size of power supplies is also increasing in order to satisfy demands for high currents. As a result, cases where there is not sufficient space to place capacitors on the board surface are increasing. To solve this problem, Panasonic Industry provides an abundant line-up of conductive polymer aluminum electrolytic capacitors (SP-Cap) with reduced height (2 mm or less) that can be placed on the back face of the board as an effective solution for the issue of implementation area which is worsening (Figure 3).

When capacitors can be placed on the surface, heat dissipation structures can be designed to be simple with high heat dissipation performance as compared to cases where tall capacitors are used, improving the reliability of hardware (Figure 4).

High Reliability

Another element prioritized in requirements for hardware for AI is long-term reliability. Especially during "learning" processes, processors continuously operate at their full capacity for long periods of time. For this reason, they are designed considering sufficient reliability so that they do not fail during operation. Requirements for the board temperature during long-term operation have become stricter than before in general as processors have increased in performance although they depend on the design of the heat dissipation structure and cooling method. Therefore, for peripheral parts, high heat resistance and long-term reliability are required. To respond to such hardware requirements, Panasonic Industry has developed conductive capacitors that can bear long-term use in high-temperature environments. They are effective options as capacitors for which reliability can be assured even when parts around the processors get hot. The figure below shows the comparison results of ESR and capacitance changes in a durability test (105℃, 2000 H) between conventional products and other companies' similar products. Although capacitors with the same capacitance and ESR specifications were selected, the results show a significant difference in voltage stabilization capacity after long-term use due to a difference in reliability. Characteristics of Panasonic Industry's conductive capacitors (SP-Cap) showed little change even after the test, indicating high reliability.