HZO ferroelectrics — wake-up, fatigue & the orthorhombic phase
Why does HZO need to “wake up,” why does it get tired, and how do we stabilize the ferroelectric phase without either?
I’m Taiqi Liu, a master’s researcher working on ferroelectric memory devices — HfO2-based HZO, FeFETs, and ferroelectric tunnel junctions — and how they can become reliable, low-power building blocks for computing-in-memory.
Short version: I make tiny capacitors remember — and then I try to make them remember reliably, for a very long time, at very low voltage.
My work sits at the intersection of materials, devices, and circuits. On the materials side, I study hafnia-based ferroelectrics — especially Zr-doped HfO2 (HZO) in both its ferroelectric and antiferroelectric phases. On the device side, I fabricate and measure MFM capacitors, ferroelectric tunnel junctions (FTJs), and ferroelectric FETs (FeFETs). On the systems side, I care about what those devices actually enable: dense non-volatile memory and computing-in-memory for edge AI.
I’m currently finishing my master’s (expected 2028) and preparing applications for Fall 2028 PhD programs in integrated circuits and electronic devices. I’m broadly interested in any IC direction where device physics still matters — emerging memory, heterogeneous integration, and hardware for AI — not just ferroelectrics.
A group where I can go from film growth to array-level demonstration — and where reliability is treated as a first-class research question, not an afterthought.
Tap any line to expand it. These are the four threads that run through almost everything I do.
Why does HZO need to “wake up,” why does it get tired, and how do we stabilize the ferroelectric phase without either?
Memory windows are easy to report and hard to keep. I work on keeping them.
Double hysteresis loops look like a problem — until you need an analog, energy-efficient synapse.
Breakdown, imprint, and fatigue are not footnotes. They decide whether a device ever leaves the lab.
Top: HZO crystal lattice over a wafer. Bottom: the FeFET stack I work with — gate, ferroelectric HZO, interlayer, and silicon channel. Illustration, not measured data.
HZO · 10 nm · TiN / Si
A compact list. Filter by type — full texts and data are available on request while several items are still under review.
La-doped HZO capacitors with an engineered TiN interface achieve wake-up-free switching with 2Pr ≈ 64 µC/cm² and endurance beyond 10⁸ cycles at 3 MV/cm.
Wafer-scale statistics linking interlayer thickness to memory window, retention, and read disturb — with a compact model that predicts array-level yield.
Double-loop HZO capacitors cycled to 10⁹ with <10% degradation, plus FTJ conductance states evaluated for analog in-memory inference accuracy.
Arrhenius analysis of FTJ retention and breakdown from 25–150 °C, proposing a screening protocol for early-life failures in CIM arrays.
Not just papers — the tools and test structures that made the papers possible.
A Python-driven probe-station workflow for PUND, endurance, and retention testing. Runs overnight cycling jobs, logs every failure, and plots publication-ready figures automatically.
A systematic wafer series — 3 thicknesses × 4 Zr ratios × 3 dopants — designed to answer one question cleanly: what actually stabilizes the orthorhombic phase in our process?
A lightweight simulator that takes measured FTJ conductance variation and predicts MNIST inference accuracy at array scale — so device targets come from system needs, not guesswork.
Thesis on reliability of HZO ferroelectric memory devices. Advanced Memory Devices Lab.
Focus on semiconductor device physics and analog IC design. Senior project: CMOS-compatible ferroelectric capacitor process.
Targeting fully-funded programs in Hong Kong, Singapore, Europe, and the US.
The practical stack behind the research above.
Languages: English (fluent, academic writing & presentation) · Mandarin Chinese (native). Outside the lab: distance running, mechanical keyboards, and keeping far too many browser tabs of arXiv papers open.
I’m especially happy to hear from prospective PhD supervisors, collaborators working on ferroelectric reliability, and anyone building CIM hardware who needs device data that’s honest about variation and failure.