In partnership with
An article published in Nature Nanotechnology on September 15 reports a wafer scale route to stacking oxide semiconductor devices above silicon.
The work concerns monolithic three dimensional integration. In this approach, device layers are formed sequentially on the same wafer rather than assembled as separate chips.
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
The researchers used atomic layer deposition to build three tiers of indium oxide devices on 200 millimeter silicon wafers. The reported process produced more than 100,000 devices.
That scale distinguishes the experiment from a demonstration containing only a few laboratory transistors. It does not, however, make the result a commercial processor.
The paper reports a threshold voltage standard deviation as low as 0.04 volts. It also reports average electron mobility reaching 91.6 square centimeters per volt second.
Threshold voltage variation describes how consistently devices switch at a chosen electrical condition. Lower variation can simplify circuit operation, but it does not by itself establish system reliability.
Electron mobility describes how readily charge carriers move through a semiconductor. The reported value is a device property, not a direct measure of an artificial intelligence workload.
The authors integrated ferroelectric and enhancement or depletion field effect devices across the stacked structure. They also demonstrated fully functional circuits that crossed between tiers.
AI made PMs faster. Multiplayer mode is still broken.
A PM can summarize research, draft a PRD, and mock up a prototype before lunch. The hard part starts when the team has to decide what actually gets built.
Jira Product Discovery gives product teams one place to capture insights, prioritize ideas with consistent frameworks, and build living roadmaps stakeholders can rally around.
And because it’s connected to Jira, the context behind every decision stays with the work—so developers and their agents know not just what to build, but why.
AI helps PMs move faster. Jira Product Discovery helps the whole team build with confidence.
Cross tier operation is a central test for vertical integration. A stack is useful only when signals can move between layers with predictable electrical behavior.
The article then evaluates a four tier compute in memory accelerator for large language model workloads. Compute in memory places selected operations near stored values, reducing some movement between memory and logic.
The reported accelerator results are modeled performance comparisons rather than measurements from a deployed commercial system. The paper reports speedups between 1.4 and 2.9 times against two dimensional baselines.
It also reports comparable improvements in energy delay product. That metric combines energy use and execution time, so a lower value represents a different tradeoff from speed alone.
The comparison depends on the paper’s device models, circuit assumptions, workload choices, and two dimensional reference designs. Those details should be examined before treating the numbers as processor benchmarks.
The study includes a custom indium oxide process design kit. Such a kit translates process and device behavior into models that circuit designers can use during early design work.
The authors present atomic layer deposition as compatible with wafer scale processing and sequential stacking. Compatibility in this context describes the proposed process path, not qualification for a manufacturing line.
The article’s data availability statement says source data are provided and additional data can be requested from the corresponding author. That arrangement supports review, but it is not the same as independent replication.
The work also leaves practical questions open. Large scale manufacturing would need evidence about yield, thermal behavior, long term drift, wiring density, and integration with established silicon processes.
The research does not show that every artificial intelligence workload benefits from vertical oxide devices. Its accelerator study is one modeled example built around large language model operations.
Its strongest evidence is therefore at the device and circuit level. The work links a material deposition method to stacked devices, cross tier circuits, and a modeled accelerator.
That chain is useful because each stage tests a different claim. Device measurements address electrical behavior. Circuit demonstrations address integration. Accelerator modeling addresses a possible systems consequence.
The paper does not establish a finished product, a production yield, or a cost advantage. It records a research platform whose next tests must connect measured hardware with larger and independently reproduced workloads.
Hire anyone, anywhere — compliant in under 3 days
Found the right hire, but no entity in their country? Remote becomes the legal employer — with contracts, benefits, and tax setup handled, plus direct access to the same in-house team that runs payroll locally.
