Samsung Electro-Mechanics says its biggest advantage in the emerging silicon capacitor (Si-Cap) market is being the only company that can supply multilayer ceramic capacitors (MLCC), silicon capacitors, and package substrates together — a turnkey bundle it argues shortens customers' development time.

Kim Won-gi, who heads the company's silicon capacitor development group, made the case at a June 11 seminar in Seoul. The company's strength, he said, is that it is fluent in semiconductor process design while also having high-frequency design capabilities, letting it respond quickly with customized solutions. That "only we can do all three" framing is the company's own pitch — but it is backed by recent commercial wins that lend it weight.

Why AI Chips Suddenly Care About Capacitors

Capacitors store electricity and release it on demand, acting as an electrical cushion that stabilizes a device's power supply. As AI spreads, the demands on components are converging on a few traits: high performance in a small size, noise isolation in densely packed designs, and stable operation at high temperatures. Samsung Electro-Mechanics' flagship here is the MLCC, which stores charge by alternately stacking ceramic dielectric and electrode layers. But because it requires stacking hundreds of layers, thickness can balloon — the gap that silicon capacitors were developed to fill.

The silicon capacitor borrows from DRAM. Fine holes are etched into a silicon wafer much as in the DRAM process, then electrodes and dielectrics are thin-film-deposited on the hole walls to form the capacitor, which is finished with wiring so it can mount on a package or substrate; the more holes, the more capacitance. "I thought it would be possible to separate just the capacitor from DRAM and build the high-performance capacitor customers want," Kim said, describing a product designed for the higher capacitance and voltage data centers require.

Thin Enough to Sit Next to the Chip

The key advantage over MLCC is thinness. A single-layer structure cuts thickness sharply — below 100 micrometers, around a tenth of a human hair — while packing more capacitance into the same area. That matters because of where the part has to go: in AI chip packages, silicon capacitors are increasingly placed around the interposer that carries the main chip and HBM, on the back of the package substrate, and even embedded inside the substrate itself, and their thin, uniform profile also helps substrate yield.

Location is the whole point. The closer a capacitor sits to the processor, the faster it can answer sudden swings in current draw, and a silicon capacitor's very low parasitic inductance and quick transient response let it damp the voltage dips and noise that erupt when an AI chip's power demand spikes — a failure mode that can corrupt computation in a densely packed package. Kim tied that growing power appetite to physical AI and AI data centers.

Complement, Not Cannibal

The company frames MLCC and Si-Cap as complementary rather than competing. Per unit of capacitance, Kim said, silicon capacitors cost more — MLCC serves the general market, while Si-Cap suits high-performance, high-precision uses — and although there had been internal worry about cannibalizing the MLCC business, in practice the two are used together. The industry expects the silicon capacitor market to grow around 18% a year through 2031. Samsung Electro-Mechanics runs the silicon-capacitor business on a fabless model, outsourcing wafer production to foundries and packaging to OSAT firms while handling design, testing, and quality itself.

From New Entrant to Named Supplier

Having begun mass production only late last year, Si-Cap is still a small share of revenue, but the company expects that to grow in earnest from this year — and recent wins back it up. In May, Samsung Electro-Mechanics disclosed a 1.557 trillion won contract, about $1 billion, to supply silicon capacitors to an undisclosed U.S. company developing next-generation AI chips, with deliveries from 2027 through 2028, and it has begun supplying Marvell. The deals push it into a market led by Japan's Murata and TSMC — which use different technical approaches, Samsung's DRAM-based and TSMC's trench-based — with newer entrants such as California's Empower Semiconductor also circling.

Kim returned repeatedly to integration as the differentiator: when a component fails, single-vendor sourcing avoids disputes over fault, and supplying MLCC, Si-Cap, and substrates together shortens customers' development cycles. Combining semiconductor process know-how with high-frequency design, he said, is the company's greatest strength — a claim that, for now, rests on SEMCO being the rare supplier that spans all three parts at once.

Frequently Asked Questions

What is a silicon capacitor?

A silicon capacitor is an ultra-thin passive component built on a silicon wafer using semiconductor processes. Fine holes are etched into the wafer, as in DRAM manufacturing, and dielectric and electrode layers are deposited on the hole walls to store charge — more holes mean more capacitance. Its single-layer structure makes it far thinner than a multilayer ceramic capacitor, so it can sit very close to a processor inside an advanced chip package.

How is a silicon capacitor different from an MLCC?

An MLCC stores charge by stacking hundreds of ceramic dielectric and electrode layers, which makes it versatile and cheap but limits how thin it can get. A silicon capacitor is thinner — below 100 micrometers — and offers faster transient response and lower parasitic inductance, which suits the high-performance, high-precision power needs of AI chips. Samsung Electro-Mechanics frames the two as complementary, with MLCC serving the general market and silicon capacitors the most demanding uses.

Why do AI chips need silicon capacitors?

AI processors draw large, rapidly changing amounts of power, which can cause voltage swings and electrical noise that disrupt computation. Placing a fast-responding capacitor very close to the chip — around the interposer, on the substrate, or embedded inside it — damps those swings at the source. Silicon capacitors are thin enough to fit in those tight locations, which is why they are increasingly used in AI accelerator packages.

Who makes silicon capacitors for AI chips?

The market has been led by Japan's Murata and Taiwan's TSMC, which use a trench-based process, with Samsung Electro-Mechanics entering using a DRAM-based approach and newer players such as California's Empower Semiconductor also competing. Samsung Electro-Mechanics has disclosed a roughly $1 billion contract with an undisclosed U.S. AI-chip developer and has begun supplying Marvell.