SOI vs bulk silicon: what is the difference?
The choice between SOI and bulk is a choice about one layer of oxide. That layer changes device physics, wafer cost and heat flow, and the industry has settled on different answers for different products.
The short answer
An SOI (silicon on insulator) wafer has a buried oxide layer (BOX) between a thin top silicon layer and the supporting substrate, so transistors are electrically isolated from the substrate; this lowers parasitic capacitance and leakage, eliminates latch-up and improves radiation tolerance. A bulk silicon wafer is a single crystal throughout, which is cheaper, conducts heat away better and is supported by the most mature manufacturing ecosystem. Most leading-edge logic and memory use bulk wafers, while RF switches, FD-SOI low-power chips and silicon photonics use SOI.
Side-by-side comparison
| Aspect | SOI | Bulk silicon |
|---|---|---|
| Structure | Three layers: substrate, buried oxide (BOX), thin top silicon | A single silicon crystal |
| Device isolation | BOX isolates devices from the substrate; trenches isolate them from each other | pn junctions and shallow trench isolation (STI) |
| Parasitic capacitance | Lower; source and drain junction capacitance is reduced | Higher |
| Latch-up | Eliminated by full dielectric isolation | Must be prevented by layout and well design |
| Radiation tolerance | Higher; less charge collected from particle strikes | Lower |
| Heat dissipation | Poorer; the BOX conducts heat badly, causing self-heating | Good; heat flows into the substrate |
| Wafer cost | Considerably higher | Lowest |
| Main manufacturing methods | Smart Cut layer transfer, bonding and thinning, SIMOX | CZ crystal growth, slicing and polishing, optional epitaxy |
| Typical uses | RF switches, FD-SOI logic, silicon photonics, MEMS, automotive and space electronics | Leading-edge logic (FinFET, GAA), DRAM, NAND, general-purpose ICs |
What the buried oxide changes
In a bulk transistor, the source and drain form pn junctions with the substrate. These junctions have capacitance that must be charged and discharged at every switching event, and they provide a leakage path. In SOI, the BOX sits directly below the devices, so junction capacitance falls sharply and the leakage path into the substrate is cut, improving speed and power at the same geometry.
Bulk CMOS also contains parasitic pnpn thyristor structures that can switch on and conduct a large current (latch-up), which designers prevent with guard rings and well taps. In SOI each device is surrounded by insulator, so latch-up cannot occur. Charge generated in the substrate by radiation is also kept away from the active layer, which is why SOI has long been used in space and aerospace electronics.
Types of SOI and where each fits
Fully depleted SOI (FD-SOI) uses a top silicon layer thin enough that the whole channel is depleted. It suppresses short-channel effects with little channel doping, and a voltage applied to the substrate under the BOX (body bias) adjusts the threshold voltage while the chip runs. It is used for low-power microcontrollers, IoT and automotive chips; STMicroelectronics and GlobalFoundries run FD-SOI processes.
RF-SOI combines a high-resistivity substrate with a charge-trap layer and is widely used for antenna switches and tuners in smartphones. Silicon photonics uses the top silicon as an optical waveguide over the oxide cladding, and MEMS uses thicker SOI as a structural layer with the oxide as an etch stop.
How SOI wafers are made
Most SOI is made by Soitec's Smart Cut process. Hydrogen ions are implanted into an oxidised donor wafer to create a weakened plane at a set depth; the donor is bonded to a handle wafer, and a heat treatment splits it along that plane. The oxide and a thin silicon film stay on the handle wafer, and the donor can be reused. The surface is then smoothed and the top silicon thickness controlled to nanometre precision.
Alternatives include SIMOX, which forms the BOX by implanting oxygen and annealing, and bonding followed by grinding and polishing for thicker layers. All of them add steps to bulk wafer production, which is why SOI wafers cost more.
Why leading-edge logic stayed on bulk
Around the 20 nm generation, planar bulk transistors reached their limits. One path was FD-SOI, keeping planar devices; the other was FinFETs on bulk silicon. Intel introduced bulk FinFETs at 22 nm, and TSMC and Samsung also built their FinFET nodes on bulk wafers. FinFETs and GAA transistors control the channel through their 3D geometry, so they suppress short-channel effects without SOI's isolation, and the cheaper bulk wafer won for mainstream leading-edge logic. FD-SOI continues as a specialised option for low-power and RF-integrated chips.
FAQ
Why are SOI wafers more expensive?
They are made from bulk wafers with extra steps: oxidation, implantation, bonding, splitting and surface finishing. Two wafers are involved in each SOI wafer (the donor can be reused), and the top silicon thickness must be held to nanometre tolerances.
What is the difference between PD-SOI and FD-SOI?
The thickness of the top silicon. In partially depleted SOI the layer is thick enough that an undepleted floating body remains under the channel, causing history effects. In fully depleted SOI the layer is so thin that it is depleted entirely, which improves short-channel control and enables body-bias tuning.
Does SOI's poor heat dissipation matter?
Silicon dioxide conducts heat roughly a hundred times worse than silicon, so transistors on SOI can self-heat. For low-power circuits this is rarely a problem, but high-current, high-performance designs must account for it.
Which Japanese companies make SOI wafers?
Shin-Etsu Handotai, part of Shin-Etsu Chemical, produces SOI wafers including Smart Cut wafers under licence from Soitec, and SUMCO also lists SOI wafers in its product range.
Terms compared
- SOI (Silicon on Insulator)A wafer with a thin silicon layer on a buried oxide, which isolates transistors from the substrate to cut capacitance and leakage.
- Bulk SiliconAn ordinary wafer made entirely of single-crystal silicon, as opposed to SOI; also the interior of a material as opposed to its surface.
- Smart CutA layer transfer technology that splits a hydrogen-implanted wafer to move a thin single-crystal layer onto another wafer; the main way to make SOI.
- Latch-upA failure in which a parasitic thyristor inside CMOS turns on and shorts the supply to ground, often destroying the chip.
Japanese companies involved
- Shin-Etsu Chemical · Silicon wafers, photoresists and photomask blanks
- SUMCO · Silicon wafers
- Toshiba Electronic Devices & Storage · Power MOSFETs, IGBTs and discrete semiconductors
- ESPEC · Environmental and reliability test chambers
More wafers and substrates comparisons
Sources
Semiconductor comparisons
- All semiconductor comparisons
- Epitaxial wafer vs polished wafer: what is the difference?
- Fabless vs foundry: what is the difference?
- Tokyo Electron vs Applied Materials: what is the difference?
- EUV vs DUV lithography: what is the difference?
- High-NA EUV vs standard EUV: what is the difference?
- Single patterning vs multi-patterning: what is the difference?
- Chemically amplified resist vs metal oxide resist: what is the difference?
Last updated: November 2, 2026