First complete 2D flash chip programs in just 20 nanoseconds

A research team in Shanghai has built a full 2D flash memory chip directly on a silicon control die. It programs in 20 nanoseconds and runs real instruction sets, not just lab demos.

The chip sits on a standard 0.13 micrometer CMOS platform and stores 1 kilobit in a NOR array. It proves that atomically thin devices can be wired into real systems with clocks, buffers, and sense amplifiers.

Why this 2D flash chip matters

Lead researcher Chunsen Liu of Fudan University led the project and is the corresponding voice for the team’s system level results. The work shows that 2D memory can move from single devices to full chips without losing speed or reliability.

Two-dimensional devices are attractive because channels only a few atoms thick switch cleanly. They also avoid many of the defects that slow down conventional interfaces.

The 2D chip matters because it clears three stubborn barriers at once. It bonds fragile layers to a rough chip surface, routes high voltages safely, and packages the stack without cooking or cracking it.

What is actually new

The team used a conformal adhesion flow that gently transfers monolayer MoS2 onto a rough CMOS top metal. The method blunts random stress that would otherwise warp an atomic layer.

A modular 3D structure separates the memory plane from the logic blocks and links them through vias. That turns device-level quirks into interface design problems that circuits can solve.

A 2D-friendly package guards every pad against static and uses low temperature wire bonding. That choice protects the atomically thin channel from heat, strain, and charge jolts.

How the 2D flash chip works

This memory chip uses NOR flash, a layout where each cell connects to the bit line in parallel, which favors fast random reads. NOR is the right match when low latency matters more than density.

Programming relies on Fowler Nordheim tunneling, a quantum tunneling process through a thin barrier under a strong field. The team matched the memory load to the CMOS drivers for clean, fast voltage waveforms.

They also designed impedance matching, the practice of sizing drivers to the load so signals settle quickly. That decision limits crosstalk across half-selected cells during program and erase steps.

What the tests showed

The full memory chip reached a 94.34 percent yield across its 1 kilobit array. It ran with 32 bit parallel writes and reads under a 5 megahertz clock with random access.

Cell programming used about 0.644 picojoule per bit based on energy per pulse. Peripheral logic drew a few milliwatts during operation, which is close to commercial standalone NOR at similar nodes.

“We believe that these system-level results represent an important milestone in extending the superiority of 2D electronics to real-world applications,” wrote Liu. The quote reflects the team’s focus on full stack integration rather than isolated devices.

The 2D flash chip followed standard command signals to store and retrieve data, much like how everyday memory in computers receives instructions. The group validated a checkerboard write across rows and measured crosstalk that stayed near zero.

How it stacks up with today’s memory

Commercial NOR typically promises around 100,000 program erase cycles in qualified parts. That level appears across vendor datasheets and remains the bar for embedded code storage.

This memory chip’s peripheral power numbers line up with that class while its cell energy looks far lower. That mix could help microcontrollers that boot from NOR and run tight power budgets.

The field also saw a separate leap this year at the device level. Earlier work reported sub-nanosecond programming in a graphene channel flash based on a new, hot carrier route.

What the results actually mean

System integration is the headline, not just faster bits. The team showed that atom-thin channels can partner with mature CMOS controls without burning yield.

The yield also clears a practical bar. Industry targets for flash manufacturing sit near 89.5 percent by roadmap accounting, and the team’s full chip tests exceeded that figure inside the small array they taped out.

Random access and instruction driven control matter more than a pretty micrograph. They show this is a real memory, not a single transistor curve.

Limits and next steps

The array is small, so scaling the memory plane will test uniformity and routing overhead. Larger planes may need new array architectures tailored to 2D physics.

Endurance at product scale needs more stress testing across corners. That includes high temperature retention, disturb, and long-program erase cycling on big arrays.

Interface circuits will need to shrink and harden for embedded use. Charge pumps, isolation rings, and level shifters must fit tight power backstops in real SoCs.

Why students should care

The project blends materials, devices, circuits, and packaging. It is a rare case where a physics trick becomes a working subsystem without hand waving.

It also shows where progress lives today. Gains come from cleaner interfaces and smarter systems as much as from new materials alone.

Context and caution

Media coverage has framed this as a turning point for 2D electronics. One concise report summarized the device-to-system leap and flagged its practical flavor.

Hype can outrun reality, but the evidence here is strong. The numbers come from full 2D flash chip tests with instruction flows and measured error rates.

The study is published in Nature.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–