Skip to content

Author

Cheng Zhuo

2 papers indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Open access Jul 2026

GTRCiM: Generic Temperature-Resilient Subthreshold-FeFET Based Compute-in-Memory

Compute-in-memory (CiM) offers a promising solution to the hardware challenges faced in artificial intelligence (AI) and the Internet of Things (IoT), particularly in tackling the ”memory wall” problem. By leveraging nonvolatile memory (NVM) devices arranged in a crossbar structure, CiM efficiently accelerates multiply-accumulate (MAC) computations, which are essential for neural networks and various AI models. Emerging NVM, i.e., Ferroelectric FET (FeFET), is a particularly appealing choice for constructing ultra-low-power CiM arrays due to its CMOS compatibility, voltage-driven write/read mechanisms and high ION/IOFF ratio. Operating FeFETs in the subthreshold region, i.e., subthreshold-FeFETs, further reduces power, but introduces vulnerability to temperature drift, leading to potential accuracy degradation. In this paper, we present subthreshold-FeFET based temperature-resilient 2T-1FeFET and 2FeFET-1T CiM cell designs that address the temperature vulnerability. These two cells leverage feedback mechanisms to minimize the output current fluctuations caused by temperature drift. We further propose GTRCiM, a novel generic temperature-resilient CiM design that accommodates both our proposed CiM cells and other typical FeFET based cells while maintaining ultra-low-power consumption with subthreshold-FeFETs. By incorporating an additional column of the CiM array as a reference, and leveraging the temperature-resilient cell design, GTRCiM compensates for the effects of temperature drift, as temperature affects both arrays uniformly. We integrate the proposed GTRCiM array with typical 1FeFET-1R, and our proposed 2T-1FeFET, 2FeFET-1T cells and validate that GTRCiM enhances the temperature resilience with our proposed cells and even the temperature vulnerable 1FeFET-1R cell. Benchmarking results at system level using the NeuroSim framework show that our proposed GTRCiM array integrated with subthreshold-FeFET based 1FeFET-1R cells achieves 28.09 TOPS/W energy efficiency, surpassing existing CiM designs while maintaining sufficient temperature resilience.

Yifei Zhou, Zeyu Yang, Meimei Ma et al. · 0 citations
Preprint Aug 2026

RODE: A Radial-Orthogonal Decoupled Engine for Optimization

Modern neural network training increasingly uses matrix-aware optimizers, yet their conditioned matrix step is typically added directly to the weight, jointly changing its norm and direction. This interaction matters because the current norm determines angular motion, while directional learning can drive norm growth and thereby alter later steps. We introduce RODE, which gives the radial and directional components separate update rules and step sizes. RODE explicitly updates the matrix Frobenius norm through a scalar radial rule, while its directional channel performs Newton--Schulz-conditioned updates in the tangent space. Controlled GPT-2 interventions show gains from both direct norm control and RODE's directional update. Across two language-modeling and two image-classification tasks, RODE outperforms both Muon variants in every direct comparison and ends with lower full-model norms. At 1.5B scale, using the learning rate transferred directly from the Qwen2-style LM sweep, RODE lowers loss from 4.145 to 3.346 and final global norm from 11964 to 2183 relative to Muon RMS, with fixed-radius RODE improving further. For Qwen3.5-9B full-parameter fine-tuning, all six optimizers use the same tuning budget and the same formal-training and evaluation settings; RODE outperforms both Muon variants on all four evaluation tasks and attains the highest mean on GSM8K and MATH-500. Thus, decoupling radial and directional dynamics offers a more effective and controllable approach to matrix optimization.

Guoxiang Xu, Bince Qu, Qi Sun et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.