Monolithic 3D CMOS inverter research explores how complementary transistors can be stacked vertically rather than placed side by side. In a 2017 demonstration, researchers built an inverter using layered semiconductor materials and reported a peak voltage gain of about 45 V/V at a 1.5 V supply. The result shows the potential of combining vertical integration with two-dimensional channel materials, while also revealing a practical circuit-level challenge: the prototype did not switch across the full supply range.
Why stack complementary devices?
Conventional CMOS logic places NMOS and PMOS transistors on the same plane. As device dimensions shrink and circuit density increases, interconnect resistance and capacitance, along with device variability, can constrain performance. Monolithic 3D integration builds active device layers sequentially, creating an opportunity to shorten average connections between circuit elements and reduce interconnect capacitance.
Vertical integration can also bring different active materials together. The demonstrated inverter used molybdenum disulfide (MoS2) for its NMOS transistor and tungsten diselenide (WSe2) for its PMOS transistor. Both are transition metal dichalcogenides (TMDs), a family of layered semiconductors. Their channel thickness can be controlled down to a few layers, a characteristic of interest for thin-body devices. The paper also discusses their dielectric properties and the potential relevance of those properties to electrostatic control as devices scale.
How the monolithic 3D CMOS inverter is configured
The NMOS and PMOS devices were fabricated on separate vertical levels, with the WSe2 PMOS positioned above the MoS2 NMOS. The transistors shared a common gate, while their source and drain electrodes remained electrically distinct. That arrangement forms the inverter’s complementary pair without requiring both channels to occupy the same plane.
The devices used 3–6 nm-thick semiconductor flakes and a zirconium dioxide (ZrO2) gate dielectric. The researchers formed the lower MoS2 transistor first, added gate and dielectric layers, and then transferred the WSe2 channel onto the stack to create the upper transistor. A forming-gas anneal followed the WSe2 transfer to help remove surface residues. This process step was associated with improved output conductance in the PMOS device, an important factor in the inverter’s measured gain.
Device characteristics and measured gain
The individual transistors showed strong on-to-off current ratios: greater than 106 for the MoS2 NMOS and about 107 for the WSe2 PMOS. Reported field-effect mobilities were 37 cm2/Vs for electrons in MoS2 and 236 cm2/Vs for holes in WSe2. These measurements describe the component devices; inverter performance depends on how their characteristics work together.
In particular, voltage gain benefits from high transconductance and high output resistance. The study linked the improved inverter gain to low output conductance in the devices, especially the WSe2 PMOS. The assembled inverter reached a peak gain of approximately 45 V/V with a 1 μm gate length and 1.5 V supply voltage. The authors reported this as the highest gain at the smallest gate length and lowest supply voltage among the monolithic 3D CMOS inverters using layered semiconductors they compared.
The key caveat: gain is not the same as usable logic
A high peak gain does not, on its own, establish that an inverter is ready to function as a logic element in a larger circuit. For reliable logic operation, the output should transition between the low and high supply levels, and the switching point should support suitable noise margins.
In this prototype, the MoS2 NMOS had a negative threshold voltage. As a result, the inverter did not switch between 0 V and the supply voltage, so the researchers could not calculate its noise margins. This is a significant qualification: the measured gain demonstrates promising electrical behavior, but the reported transfer characteristics did not establish full rail-to-rail logic operation.
Threshold-voltage tuning
The study examined substrate back-biasing as one way to adjust the NMOS threshold voltage and shift the inverter’s switching point. Applying a more negative back bias made the NMOS threshold voltage less negative. In the reported measurements, gain increased from 26 V/V at a −50 V back bias to 31 V/V at −60 V and 45 V/V at −70 V. The switching point also moved toward more positive input voltages.
Other approaches identified by the researchers include gate work-function engineering and channel doping. These are potential methods for bringing transistor thresholds into a range that supports the desired inverter transfer characteristics; the paper does not establish that the demonstrated device had already achieved the required noise margins.
What the result means for 3D integration
For engineers and technical decision-makers, the demonstration is best understood as a device-and-architecture milestone, not an industrial automation component or a production-ready logic platform. It illustrates how different layered semiconductors can be integrated into vertically stacked complementary transistors and how process details can influence circuit-level measurements.
The practical lesson is to assess 3D CMOS research at both the transistor and circuit levels. Mobility, on-to-off ratio, output conductance, gain, switching range, and noise margins answer different questions. The 45 V/V result is notable, but realizing functional logic also requires threshold-voltage control and a suitable voltage transfer characteristic. Continued progress in those areas will determine how effectively layered-semiconductor integration can translate from a laboratory inverter to more complete circuits.