Hybrid MoS2 and Carbon-Nanotube CMOS Ran a 4-Bit Adder at Sub-Picowatt Static Power
Researchers integrated two-dimensional molybdenum disulfide n-type transistors with one-dimensional aligned carbon-nanotube p-type transistors. The platform produced inverters, logic gates, a 4-bit full adder and compact SRAM, while reporting less than 1 pW peak static power and a 19 ps single-stage delay.
Quick summary
- The circuit combines MoS2 n-type transistors with aligned carbon-nanotube p-type transistors.
- It demonstrated inverters, four basic logic gates, a 4-bit full adder and a six-transistor SRAM cell.
- The best reported inverter metrics were under 1 pW peak static power and a 19 ps single-stage delay.
- This is a laboratory platform, not a manufacturing-ready replacement for silicon CMOS.
A team led from Nanjing University has built complementary logic by assigning the two halves of a CMOS circuit to different low-dimensional semiconductors. Single-crystal molybdenum disulfide, or MoS2, supplies the n-type transistors, while aligned carbon nanotubes supply the p-type devices. The pairing addresses a recurring problem in post-silicon electronics: an excellent transistor material in one polarity does not automatically provide an equally strong partner of the opposite polarity.
The work, published in Nature Electronics on 16 November 2026, goes beyond an isolated transistor. The researchers fabricated enhancement-mode top-gate devices, arrays of complementary inverters, NAND, NOR, AND and OR gates, and a cascaded 4-bit full adder. They also demonstrated a six-transistor static random-access memory cell with an area equivalent to a 180 nm technology node.
The important result is not simply that MoS2 and nanotubes can share a chip. It is that their electrical behavior was balanced well enough to build several layers of working digital logic.
Why mix one-dimensional and two-dimensional materials?
CMOS logic uses a p-type transistor to pull an output high and an n-type transistor to pull it low. Ideally, both devices switch at comparable voltages and drive similar currents. Low-dimensional materials can keep electrostatic control as channels become thin, but many naturally favor one carrier type. The new platform uses each material where it performs best rather than forcing one material to do both jobs.
A hard-mask process and annealing before dielectric deposition helped the team make top-gated transistors with roughly symmetric threshold voltages. The resulting inverters operated from rail to rail. The paper reports peak static power below 1 pW and an estimated single-stage delay of 19 ps. Uniform device arrays then supported the multi-stage adder, where carry signals propagated through four connected stages.
The memory demonstration matters because SRAM is more demanding than a single logic gate. The 3.96 square micrometre cell completed read and write operations, with a hold noise margin equal to 22% of the operating voltage and a read noise margin of 40%. These figures show functional integration, although they do not yet establish commercial yield or long-term reliability.
What the experiment does not prove
Reality check
The devices were fabricated in a research environment and compared mainly with other emerging-material circuits. The paper does not report high-volume wafer yield, years-long stability, foundry-compatible process costs or direct system-level superiority over advanced silicon. The 180 nm figure is an area equivalence for the SRAM design, not evidence that the full process is ready at a commercial node.
The static-power and delay values are promising, but they come from specific test structures and operating conditions. Contact resistance, nanotube alignment, material transfer, contamination and device-to-device variation remain practical barriers. Scaling a few functional blocks into millions of reliable transistors is a different challenge from demonstrating a 4-bit adder.
Still, the work gives circuit designers a more complete test of mixed-dimensional integration. It complements NewTqnia's coverage of a WSe2 photodiode that received visible-light data at 10 Gbit/s and neuromorphic hardware used in a closed temperature-control loop. Together, these studies show how atomically thin and unconventional devices are moving from component measurements toward small but functional systems.
The next decisive evidence will come from larger circuit counts, statistically reported yields, endurance testing and fabrication flows that minimize transfers between separate material stacks. Until then, mixed MoS2 and carbon-nanotube CMOS is best read as a credible circuit-level research platform, not an imminent silicon replacement.
Verified topics and entities
Sources and citations3 sources
External references used to support the reporting in this article.
Published by
NewTqnia Science Desk
An institutional editorial team within NewTqnia