The Material That Can Both Compute and Remember - and May Replace Silicon

Silicon has had a remarkable run. For sixty years, the semiconductor industry has shrunk transistors by roughly half every two years, packing more computing power into less space while keeping energy costs manageable. That trajectory is ending. At sub-nanometer scales, silicon transistors leak current, overheat, and consume disproportionate power just to function. The question of what comes next is not academic - it is the central engineering problem of the AI era.

A team led by Professor Seunguk Song at Sungkyunkwan University (SKKU) in South Korea, working with colleagues at the Institute for Basic Science, the University of Pennsylvania, and the U.S. Air Force Research Laboratory, has published a comprehensive roadmap in Nature Reviews Electrical Engineering for one of the most promising candidates: indium selenide, or InSe, a semiconductor that can be peeled down to layers just a few atoms thick.

What makes indium selenide different

InSe belongs to the family of two-dimensional materials - solids where electronic behavior changes dramatically as thickness approaches the atomic scale. What distinguishes it from other candidates, including the widely studied molybdenum disulfide, is a combination of properties that happen to align well with the requirements of next-generation computing.

Its electrons have an extremely small effective mass, meaning they accelerate rapidly under an applied voltage and travel through the material with minimal resistance - a property known as ballistic transport. High-speed operation at low applied voltage translates directly to lower power consumption. For AI inference chips that run continuously in data centers consuming gigawatts of electricity, that efficiency gap matters enormously.

Equally important is that InSe's electronic properties change depending on how its atomic layers are stacked. Certain arrangements produce ferroelectric behavior - the material retains a memory of its electrical state after the power is removed, similar to how a magnet retains its orientation. A material that can both compute (logic) and remember (non-volatile storage) within a single device is a significant departure from conventional chip architecture.

Beyond Von Neumann

Modern computers shuttle data constantly between processors and memory - a bottleneck known as the Von Neumann bottleneck that becomes increasingly severe as datasets grow larger and AI models more complex. A substantial fraction of the energy consumed by a large language model goes not into computation but into moving data back and forth across that gap.

InSe-based devices could perform computation and store results in the same physical location, a design philosophy called in-memory computing. The roadmap describes a specific path from proof-of-concept quantum transistors to scalable non-volatile memory systems - not a vague promise but a technology trajectory with defined milestones.

"This research is significant because Indium Selenide is not just a new material; it represents a shift in the computing paradigm," Song said. "We expect it to evolve into a core platform that bridges quantum information technology with low-power semiconductor engineering."

The obstacles that remain

The roadmap is honest about what stands between laboratory results and industrial adoption. Growing InSe over large areas with the atomic-level uniformity that chip fabrication requires is still an unsolved manufacturing challenge. The material also degrades in the presence of oxygen and moisture - oxidation stability is a known problem that will require either protective encapsulation or processing in controlled atmospheres.

Both challenges are primarily engineering problems rather than fundamental physics barriers, which is part of why the roadmap format is appropriate. The authors are not claiming that InSe works at scale; they are describing the technical path that would need to be cleared for it to do so.

The applications they point toward - quantum computer peripherals and ultra-low-power AI semiconductors - are both high-value targets with strong commercial and defense interest, which helps explain why the research attracted support from the U.S. National Science Foundation, the Office of Naval Research, and the Air Force Office of Scientific Research alongside Korean government funding.

Where InSe fits in the competitive landscape

InSe is not the only two-dimensional semiconductor being developed as a silicon alternative. Graphene, transition metal dichalcogenides, and boron nitride all have active research communities. What the SKKU roadmap contributes is a systematic evaluation of InSe's specific advantages in both logic and memory applications, and a credible timeline for addressing the remaining barriers - the kind of document that tends to attract industry attention and funding for the next phase of development.

Silicon will not vanish from chips any time soon. But the next generation of computing - the one that runs AI without requiring a power plant - will almost certainly be built on something else. InSe is a serious candidate for part of that future.