Murata Silicon Capacitors: AI Optical Module Power

1. What Happened

The speed ramp in Ethernet is hard to overstate. In the year 2000, 10 gigabit per second was the frontier. Today, 1.6 terabit per second (1.6T) is the mainstream high-end of optical modules, and the industry is already charting the move to 3.2T around 2028, with some observers pointing to 2030. That is three orders of magnitude of speed growth in roughly two decades, and it has turned the passive components inside every optical module into a design battleground.

At the center of that battleground is the humble capacitor — specifically, the silicon capacitor. Unlike the ceramic capacitors that have dominated electronics for decades, silicon capacitors are built using semiconductor processes, which gives them a fundamentally wider and flatter frequency response. As optical modules push toward single-lane 200G and eventually 400G and 800G, the capacitors that couple and decouple signals have to keep up across an ever-wider bandwidth. That is exactly the capability Murata is selling.

At the 2026 China International Optoelectronic Exposition, Murata's silicon-capacitor product line presented a full portfolio aimed at AI data-center optical interconnect. Olivier Gaborieau, senior product line manager for silicon capacitors at Murata, was blunt about what is driving the ramp: "In my view, AI is the biggest factor. In 2014 or 2016, many of the things we see now simply did not exist, because the arrival of AI has been the biggest driver pushing the entire module industry to higher speeds."

The product lineup Murata showed spans ceramic capacitors, silicon capacitors, power ICs, crystal oscillators and thermistors, inductors, filters, thin-film lithium niobate (TFLN) modulators, and LTCC and LCP substrates — a broad portfolio aimed squarely at the next generation of pluggable optical modules. But the silicon capacitor is the piece of that story that connects directly to the industry's hardest technical problem: keeping signal integrity intact as speeds climb.

2. Why It Matters

For anyone sourcing components for optical interconnect, the silicon-capacitor story matters because it sits at the intersection of three forces that are reshaping the entire data-center supply chain: speed, integration, and miniaturization. Optical modules are getting faster at the same time they are getting smaller and more complex, and the passive components inside them are being asked to do more in less space. That is a recipe for supply-chain tension, and it is why a single component category — the capacitor — has become a strategic talking point.

The speed force is the most obvious. As optical modules move from 100G-per-lane to 200G-per-lane and beyond, the bandwidth that capacitors must handle keeps widening. A capacitor that worked fine at 100G can become a signal-integrity bottleneck at 200G. The result is that module makers are forced to re-evaluate the passives they have used for years, and suppliers with the right high-frequency performance gain pricing power and design influence.

The integration force is subtler but arguably more important. Optical modules are migrating from pluggable form factors — where the module is a separate box that plugs into a switch — toward non-pluggable architectures such as NPO (near-package optics) and CPO (co-packaged optics). In CPO, the optics are co-packaged directly with the switch silicon, which shortens the electrical path, removes much of the DSP, and cuts power. Murata cites a comparison in which a 3.2T CPO on-board optical module achieves more than 100-fold volume reduction versus a 4×OSFP800 pluggable module. That kind of density changes what components can fit, and where.

The miniaturization force is the consequence. As modules shrink and protocols demand more capability, the number of passive components inside each module is actually rising, not falling. That creates a physical contradiction: more passives, less space. Murata's answer is embedded capacitors — passives built into the substrate itself rather than soldered on top — and custom silicon substrates. For buyers, this is the leading edge of a broader shift we have tracked across the chip supply chain: capacity and integration constraints that look distant today tend to arrive faster than expected, which is a theme we explored in our look at the AI optical module bottleneck.

To put the demand in perspective, consider what an AI cluster looks like at the network layer. A single hyperscale training cluster can connect tens of thousands of accelerators through a hierarchy of switches and optical links, and the optical modules that carry those links are counted in the hundreds of thousands across a single deployment. Every one of those modules carries the same economics: more bandwidth per module, less power per bit, and less space per rack. When a component change saves even a small amount of power or space per module, the aggregate savings across a cluster are enormous. That is why the passive components inside the module — the capacitors in particular — have become the subject of serious engineering attention rather than an afterthought, and it is why we have been tracking the broader supply-chain implications in our coverage of the signals shaping the 2026 chip supply chain.

3. The Technical Background

3.1 The speed ramp, in context

To understand why a capacitor matters at all, it helps to understand what "1.6T" and "3.2T" actually mean at the component level. A 1.6T optical module typically combines multiple lanes — for example, eight lanes of 200G each, or sixteen lanes of 100G. Each lane carries a high-speed serial signal, and each lane's signal path includes coupling capacitors that pass the AC signal while blocking DC, and decoupling capacitors that smooth the power supply. As the per-lane rate rises, the frequencies that those capacitors must pass rise with it, and the parasitic inductance and resistance of the capacitor itself start to eat into the signal.

This is where the difference between ceramic and silicon capacitors becomes decisive. A ceramic capacitor is a discrete component with a limited frequency response; beyond a certain point its parasitic inductance makes it behave less like a capacitor and more like an inductor, which is exactly what a high-speed signal path does not want. A silicon capacitor, by contrast, is fabricated on a semiconductor wafer with very tight dimensional control, which lets it maintain a flat, wideband response into the tens and even hundreds of gigahertz. That is the "wideband" in wideband silicon capacitor.

The optical module itself has an evolution that parallels the speed ramp. In the early 2000s, the standard form factors were relatively bulky pluggable modules such as GBIC and SFP, each carrying a single lane. As speeds climbed, the industry moved to QSFP and then QSFP-DD and OSFP, which pack multiple lanes into a single module to reach 400G, 800G, and now 1.6T. Each generation of form factor has pushed more bandwidth through a smaller package, and each has forced the supporting components — including the capacitors — to shrink while performing at higher frequencies. The silicon capacitor is a direct response to that pressure.

The technical core of the story is the difference between a semiconductor process and ceramic stacking. A ceramic capacitor is a multi-layer stack of ceramic dielectric and metal electrodes, and its useful frequency range is limited by the inductance of that structure. A silicon capacitor is fabricated the way a chip is, with lithographically defined features that can be made far smaller and more uniform. The result is a capacitor whose impedance stays low and flat to frequencies no ceramic part can reach — which is precisely why the 220 GHz figure Murata cites is possible at all.

3.2 Pluggable, NPO, and CPO: three ways to pack optics

The other half of the story is the physical architecture of the optical module itself. In a traditional pluggable module, the optics sit in a separate, replaceable box connected to the switch ASIC by a relatively long electrical path. As speeds rise, that long path becomes expensive in power and signal integrity, which is why the industry is moving toward tighter integration.

NPO, or near-package optics, shortens that path by placing the optics closer to the switch package, reducing the need for some DSP functions and cutting power. CPO, or co-packaged optics, goes further, integrating the optics directly with the switch silicon, either in a 2.5D configuration — where the optical and electrical dies sit side by side on a common substrate — or a 3D configuration, where they are stacked. CPO is widely described as the endgame of optical integration, and it is the form factor where the 100-fold volume reduction comes from.

Murata's forecast is that about 20% of 1.6T and 3.2T modules will use non-pluggable architectures, driven by the scale-networking requirements of the largest AI clusters. That is not a niche; it is a meaningful slice of a market measured in the billions of dollars, and it is the slice where component size and integration matter most.

3.3 TFLN and the modulator side

The silicon capacitor is only one part of the optical module's bill of materials. On the modulator side, thin-film lithium niobate (TFLN) has emerged as a key technology for the electro-optic modulators used in 2T-and-above pluggable modules, and it also supports CPO and NPO architectures. Murata showed TFLN modulators alongside its silicon capacitors, reflecting a strategy of covering both the signal-coupling and the modulation functions of a high-speed optical link.

The relevance for buyers is that the highest-speed optical modules now pull together a cluster of specialized technologies — wideband passives, TFLN modulators, high-frequency substrates — that only a handful of suppliers can provide. That concentration, in turn, means that the design choices made early in an optical-module program tend to lock in specific suppliers and specific components, which raises the stakes of getting the passive-component selection right the first time.

Market analysts see the same transition. Yole and other research firms have charted the shift of optical interconnect toward higher speeds and tighter integration, and the silicon-capacitor opportunity sits squarely within that forecast. What matters for a buyer is not the exact market number but the direction: the components that enable the next speed tier are becoming specialized enough that early design-in matters, and late movers face a smaller set of qualified choices.

3.4 Why embedded capacitors are the next step

The most forward-looking part of Murata's pitch is the embedded capacitor. In a traditional design, capacitors are discrete components soldered onto the module board. In an embedded design, the capacitance is built into the substrate itself, which saves board space and shortens the electrical path even further. As module sizes shrink while component counts rise, embedded capacitance moves from a niche technique to a mainstream option.

Murata is supporting this shift with custom silicon substrates and, for NPO and CPO 2.5D packages, pre-soldered ball solutions. The company has also built modeling tools that let customers evaluate embedded-capacitor performance early in the design phase. For a procurement team, the takeaway is that the bill of materials for a next-generation optical module is becoming less about buying discrete parts and more about specifying integrated, custom solutions — which changes how second-sourcing and supplier relationships work.

It is worth running the counterfactual to see why this matters. If capacitors stayed purely discrete, the shrinking space inside CPO and NPO modules would eventually force module makers to choose between reducing component count and sacrificing performance — a trade nobody wants. The embedded-capacitor path resolves that tension by moving the passive into the substrate, but it also means the capacitor stops being a commodity that can be swapped between suppliers at will. Once a design commits to a custom silicon substrate, the supplier relationship deepens, and the cost of switching rises. That is a structural shift in how optical-module passives are sourced, not just a technology upgrade.

4. The Numbers

The following table summarizes the key performance and market figures Murata shared around its silicon-capacitor portfolio. These numbers matter because they are the parameters that determine whether a capacitor can actually keep up with a 200G or 400G lane.

The 220 GHz figure is the headline. It means the capacitor still behaves like a capacitor at a frequency far beyond what a ceramic part can reach, which is precisely what a 200G-per-lane signal needs. The insertion-loss and return-loss figures describe how transparent the capacitor is in the signal path — lower insertion loss and higher return loss both mean less degradation of the very high-speed signal the module is trying to carry. The 0.005 dB/GHz figure is a way of saying the response is flat, not peaked, across the operating band.

The space and power figures — up to 70% space saving and roughly 80% power saving — are the commercial arguments that get a design engineer's attention. In a CPO module where every cubic millimeter and every watt count, a component that saves both space and power is worth a premium, which is why Murata can position silicon capacitors as more than a commodity. The 20-year lifetime figure matters for infrastructure buyers who expect optical equipment to stay in service for a long time.

5. Who Is Affected

Optical module makers are the most direct audience. Companies building 800G, 1.6T, and eventually 3.2T modules are the ones choosing between ceramic and silicon capacitors, and between discrete and embedded passives. Their design decisions determine whose components get designed in, and those decisions tend to be sticky across a product generation.

Data-center operators and hyperscalers are the indirect but ultimate beneficiaries. AI clusters are the demand engine behind the speed ramp, and the operators building those clusters care about power, density, and reliability — all of which the silicon-capacitor story touches. A component that saves power and space in every module compounds across a data center with hundreds of thousands of modules.

Component buyers and procurement teams should read this as a leading indicator. When a supplier like Murata says the passive-component count inside a shrinking module is rising, and that embedded passives are the answer, it signals that the bill of materials for optical modules is about to shift. Buyers who have been sourcing discrete capacitors for years may find their next designs call for custom substrates and integrated solutions, which changes the sourcing playbook.

Competing passive-component suppliers — the ceramic-capacitor leaders in particular — face a technology transition. Ceramic capacitors are not going away, but the highest-speed, most space-constrained optical modules are exactly where silicon capacitors and embedded passives are making inroads. That pressure will shape the passive-component market over the next several years.

There is also a corporate context worth understanding. Murata is best known as the world's largest maker of ceramic capacitors — the commodity passives found in virtually every electronic device — and its silicon-capacitor line is a deliberate step into a more specialized, higher-value tier built on semiconductor fabrication rather than ceramic stacking. For Murata, that is a way to differentiate from the ceramic-capacitor competition and capture the premium end of the optical-module market. For buyers, it means a supplier that already has deep relationships across the electronics industry is now pushing into the specific, high-frequency passives that next-generation optical modules require — which both broadens the sourcing options and concentrates expertise in a company that already spans much of the passive-component supply chain.

6. Scenarios

Scenario A — Steady adoption: silicon capacitors ride the 1.6T wave

The base case is that silicon capacitors continue to win sockets in the highest-speed modules as the industry works through the 1.6T ramp and into 3.2T. Ceramic capacitors remain dominant in lower-speed and cost-sensitive designs, while silicon and embedded passives take the premium, space-constrained tier. For buyers, this is a gradual shift with no sudden disruption — but one that favors getting familiar with silicon-capacitor suppliers early.

Scenario B — Embedded passives go mainstream

A more aggressive case is that embedded capacitance becomes the default for CPO and NPO modules faster than expected, driven by the physical limits of squeezing more discrete components into shrinking modules. If that happens, the discrete passive business for the highest-speed modules shrinks in favor of custom substrates, and the balance of power shifts toward suppliers that can do both the substrate and the component. Buyers would need to move from sourcing parts to specifying integrated solutions.

Scenario C — A capacity squeeze emerges later

The risk case is that the ramp to 3.2T collides with a capacity constraint. Murata says there is no bottleneck today, and with both 6-inch and 8-inch lines running that is credible — but silicon capacitors are a semiconductor product, and semiconductor capacity has a way of tightening when demand accelerates. If AI-driven optical demand outpaces silicon-capacitor capacity, lead times would stretch for exactly the components the fastest modules need. Buyers who lock in supply early would be insulated; those who treat silicon capacitors as an afterthought would not.

In practice the outcome will be a blend: steady adoption now, a gradual shift toward embedded passives, and a capacity question that becomes worth watching as 3.2T approaches. The common thread is that a component category most buyers never think about — the capacitor — is becoming a strategic input to the AI networking build-out.

The broader lesson is one of supplier strategy. Murata is not just selling a component; it is selling an architecture — a way of moving passives into the substrate and into custom solutions so that the module can keep shrinking while the lane count keeps rising. That is the same playbook seen elsewhere in the AI supply chain, where the winning suppliers are the ones that can bundle more of the solution. For a buyer, the implication is that the passive-component decision is becoming a design decision, and design decisions are made early and are hard to reverse.

7. What Buyers Should Do

First, recognize that the passive components in optical modules are changing. If your bill of materials includes optical modules or the components inside them, the capacitor is no longer a commodity afterthought. The highest-speed designs are moving to silicon and embedded passives, and that shift will reach more designs over time.

Second, ask your module suppliers about their passive-component strategy. For a 1.6T or 3.2T module, find out whether they are using silicon capacitors, whether they plan to move to embedded capacitance, and how they are managing the space constraints of NPO and CPO. That conversation surfaces the design risk early.

Third, track the silicon-capacitor supply base. Murata is a leader, but it is not alone in the silicon-capacitor space, and the category is small enough that capacity can tighten quickly when AI demand accelerates. Knowing who the qualified suppliers are — and having a second source — matters more for a specialized part than for a commodity.

Fourth, watch the 3.2T transition as a timing signal. The move to 3.2T around 2028 is the next major inflection, and it will drive a wave of design changes across the optical supply chain. Buyers who plan for that transition now — by engaging suppliers, qualifying alternatives, and building buffer where lead times could stretch — will be positioned ahead of the crowd.

Fifth, do not assume the capacitor decision is someone else's problem. In a traditional bill of materials, capacitors are often left to the module supplier or the contract manufacturer to choose, and the buyer never sees them. But in a 1.6T or 3.2T module, the choice of capacitor — ceramic versus silicon, discrete versus embedded — directly affects the module's power, size, and reliability, which are exactly the things a buyer cares about. Asking a simple question — "what passives are you using, and have you qualified a second source?" — surfaces the design risk before it becomes a supply problem.

If you are sourcing optical modules, silicon capacitors, or the surrounding passive and active components, our team can help you map suppliers and confirm availability against your bill of materials.

8. Timeline

• 2000 — Ethernet speeds reach 10 gigabit per second.
• 2014–2016 — Early AI-era optics; many of today's high-speed module features did not yet exist.
• 2026 — 800G and 1.6T modules ship in growing volumes; Murata showcases its silicon-capacitor portfolio for AI optical interconnect.
• 2028 — 3.2T optical modules expected to enter the market.
• 2028–2030 — Non-pluggable architectures (NPO/CPO) gain share; embedded capacitors move toward the mainstream.

9. FAQ

Q: What is a silicon capacitor, and how is it different from a ceramic capacitor?
A: A silicon capacitor is fabricated using semiconductor processes, giving it a wider, flatter frequency response than a discrete ceramic capacitor. That makes it better suited to the very high frequencies of 200G- and 400G-per-lane optical signals.

Q: Why do optical modules need wideband capacitors?
A: Each high-speed lane has coupling capacitors that pass the AC signal and decoupling capacitors that smooth the power supply. As per-lane rates rise, the capacitors must operate across a wider bandwidth without degrading the signal.

Q: What are NPO and CPO?
A: NPO (near-package optics) and CPO (co-packaged optics) are non-pluggable ways to integrate optics with the switch silicon. CPO integrates them directly, achieving large reductions in size and power compared with pluggable modules.

Q: What is TFLN?
A: Thin-film lithium niobate is a material used in electro-optic modulators for high-speed optical modules, including 2T-and-above pluggable designs and CPO and NPO architectures.

Q: What is an embedded capacitor?
A: An embedded capacitor is built into the substrate rather than soldered on as a discrete part. It saves board space and shortens the electrical path, which is increasingly important as modules shrink while component counts rise.

Q: Is there a supply risk for silicon capacitors?
A: Murata reports no capacity bottleneck today, running both 6-inch and 8-inch lines. However, as a semiconductor product, silicon-capacitor capacity could tighten if AI-driven optical demand accelerates, so it is worth tracking.

Q: How does the 3.2T transition affect component sourcing?
A: The move to 3.2T around 2028 will drive a wave of optical-module redesigns, and each redesign reopens the passive-component selection. Buyers should expect suppliers to push silicon and embedded capacitors into those designs, and should engage early to secure qualified supply before the transition tightens capacity.

Q: Are silicon capacitors a drop-in replacement for ceramic capacitors?
A: Not in general. Silicon capacitors are a specialized, higher-performance option for high-frequency, space-constrained applications; ceramic capacitors remain the right choice for many lower-speed and cost-sensitive designs. The two technologies are complementary rather than interchangeable, which is why design selection matters.

10. Glossary

Optical module

The transceiver that converts electrical signals to optical signals and back, used to connect switches and servers in data centers.

1.6T / 3.2T

Optical module data rates of 1.6 and 3.2 terabits per second.

Silicon capacitor

A capacitor fabricated with semiconductor processes, offering wideband high-frequency performance.

AC coupling

Passing the AC signal while blocking DC, typically done with a series capacitor.

DC decoupling

Smoothing the power supply by placing capacitors across the power rail.

NPO

Near-package optics, an intermediate step between pluggable modules and full co-packaging.

CPO

Co-packaged optics, where optics are integrated directly with the switch silicon.

TFLN

Thin-film lithium niobate, used in high-speed electro-optic modulators.

Embedded capacitor

A capacitor built into the substrate rather than mounted as a discrete component.

Insertion loss / return loss

Measures of how much signal a component absorbs or reflects; lower insertion loss and higher return loss are better.

Ceramic capacitor

A capacitor built from stacked ceramic layers, the dominant discrete passive component in electronics.

QSFP-DD / OSFP

Multi-lane pluggable optical module form factors used for 400G, 800G, and 1.6T links.

Pluggable module

A removable optical transceiver that plugs into a switch or server, as opposed to co-packaged optics.

11. Sources

• Murata Manufacturing — silicon-capacitor product line materials and statements from the 2026 China International Optoelectronic Exposition.
• Interview with Olivier Gaborieau, senior product line manager, Murata, as reported by International Electronic Business (ESMC).
• Yole Group — optical module and RF market data.

The silicon capacitor is a small component with an outsized role in what comes next for optical networking. As Ethernet climbs from 1.6T to 3.2T and AI clusters keep demanding more bandwidth, the passives that keep those signals clean are moving from the background to the center of the design conversation. For buyers, the lesson is the same one that applies across the semiconductor supply chain: the parts you never had to think about are often the ones that become the constraint first. Murata's bet on silicon capacitors is, in that sense, a bet on where the next bottleneck will form — and it is a reminder that the components that quietly disappear into the substrate today are the ones that will matter most when the next speed tier arrives. For procurement teams, that is the whole story in miniature: the component nobody was watching is often the one that determines whether a product ships on time, and the teams that start watching early are the ones that stay ahead of the constraint.