arXiv subjects

Tian Pei

Publications and source records attributed to Tian Pei.

8 recordsLinked to original sources

arXiv2026

High-Fidelity Hole Spin Qubits Reveal Quadrupolar Nuclear-Bath Dynamics in Isotopically Purified Planar Germanium

Planar Germanium has emerged as a promising platform to build spin-based large scale quantum computers. By exploiting the anisotropic hyperfine interaction of holes in Ge, qubits with long T2* have been recently realized. While the performance of single qubits is still more or less limited by 73Ge nuclear spin fluctuations, the site-to-site variation of qubit sweet spot becomes obstacles to maintaining high fidelity of each qubit across the whole wafer. To achieve high performance Ge-based quantum circuit, it is therefore essential to eliminate the origin source of hyperfine noise. In its Silicon counterparts, reduction of 29Si abundance enables exceptional high-fidelity operation. In contrast, hole qubits based on isotopically purified Ge have not been demonstrated. Here, we report the synthesis of high quality 2-dimensional hole gas (2DHG) with enriched 70GeH4 precursor. Due to the suppression of nonzero spin nucleus, the qubits' T2* on the sweet spot is moderately extended beyond 20 us, surpassing the previous best reported Ge hole qubits. More importantly, the qubits' T2* off the sweet spot is enhanced to above 3 us, enabling single qubit gate fidelity exceeding 99.9% in both operating regimes. Hahn-echo spectroscopy further resolves a finite-frequency nuclear-noise channel that is distinct from the conventional Larmor-linked hyperfine response. We associate this channel with quadrupole-modified dynamics of residual 73Ge nuclei sampling local electric-field gradients near the Ge/SiGe interface. Its field scaling and angle-dependent visibility are consistent with a qubit-visible quadrupolar nuclear-noise component transduced through the anisotropic hyperfine interaction of Ge holes. These results establish isotopically purified planar Ge as a high-coherence scalable platform for hole spin qubits and provide a spectroscopic probe of interfacial quadrupolar nuclear dynamics.

quant-ph↗

arXiv2025

Tunable Wigner Molecules in a Germanium Quantum Dot

The interplay between Coulomb interactions and kinetic energy underlies many exotic phases in condensed matter physics. In a two-dimensional electronic system, If Coulomb interaction dominates over kinetic energy, electrons condense into a crystalline phase which is referred as Wigner crystal. This ordered state manifests as Wigner molecule for few electrons at the microscopic scale. Observation of Wigner molecules has been reported in quantum dot and moire superlattice systems. Here we demonstrate hole Wigner molecules can be formed in a gate-defined germanium quantum dot with high tunability. By varying voltages applied to the quantum dot device, we can precisely tune the hole density by either changing the hole occupancy or the quantum dot size. For densities smaller than a certain critical value, Coulomb interaction localizes individual holes into ordered lattice sites, forming a Wigner molecule. By increasing the densities, melting process from a Wigner molecule to Fermi liquid-like particles is observed. An intermediate configuration which indicates the coexistence of ordered structure and disordered structure can be formed within a narrow effective density range. Our results provide a new platform for further exploration of the microscopic feature of strong correlated physics and open an avenue to exploit the application of Wigner molecules for quantum information in a very promising spin qubit platform.

cond-mat.mes-hall↗

arXiv2024

DeepSeek-V3 Technical Report

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.

cs.CL↗

arXiv2024

Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning

The rapid progress in Deep Learning (DL) and Large Language Models (LLMs) has exponentially increased demands of computational power and bandwidth. This, combined with the high costs of faster computing chips and interconnects, has significantly inflated High Performance Computing (HPC) construction costs. To address these challenges, we introduce the Fire-Flyer AI-HPC architecture, a synergistic hardware-software co-design framework and its best practices. For DL training, we deployed the Fire-Flyer 2 with 10,000 PCIe A100 GPUs, achieved performance approximating the DGX-A100 while reducing costs by half and energy consumption by 40%. We specifically engineered HFReduce to accelerate allreduce communication and implemented numerous measures to keep our Computation-Storage Integrated Network congestion-free. Through our software stack, including HaiScale, 3FS, and HAI-Platform, we achieved substantial scalability by overlapping computation and communication. Our system-oriented experience from DL training provides valuable insights to drive future advancements in AI-HPC.

cs.DC↗

arXiv2024

DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model

We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models.

cs.CL↗

arXiv2024

DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

The rapid development of open-source large language models (LLMs) has been truly remarkable. However, the scaling law described in previous literature presents varying conclusions, which casts a dark cloud over scaling LLMs. We delve into the study of scaling laws and present our distinctive findings that facilitate scaling of large scale models in two commonly used open-source configurations, 7B and 67B. Guided by the scaling laws, we introduce DeepSeek LLM, a project dedicated to advancing open-source language models with a long-term perspective. To support the pre-training phase, we have developed a dataset that currently consists of 2 trillion tokens and is continuously expanding. We further conduct supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) on DeepSeek LLM Base models, resulting in the creation of DeepSeek Chat models. Our evaluation results demonstrate that DeepSeek LLM 67B surpasses LLaMA-2 70B on various benchmarks, particularly in the domains of code, mathematics, and reasoning. Furthermore, open-ended evaluations reveal that DeepSeek LLM 67B Chat exhibits superior performance compared to GPT-3.5.

cs.CL↗

arXiv2016

Resonant optomechanics with a vibrating carbon nanotube and a radio-frequency cavity

In an optomechanical setup, the coupling between cavity and resonator can be increased by tuning them to the same frequency. We study this interaction between a carbon nanotube resonator and a radio-frequency circuit. In this resonant regime, the vacuum optomechanical coupling is enhanced by the DC voltage coupling the cavity and the mechanical resonator. Using the cavity to detect the nanotube's motion, we observe and simulate interference between mechanical and electrical oscillations. We measure the mechanical ring-down and show that further improvements to the system could enable measurement of mechanical motion at the quantum limit.

cond-mat.mes-hall↗

arXiv2016

Photon-assisted tunneling and charge dephasing in a carbon nanotube double quantum dot

We report microwave-driven photon-assisted tunneling in a suspended carbon nanotube double quantum dot. From the resonant linewidth at a temperature of 13 mK, the charge dephasing time is determined to be 280 +- 30 ps. The linewidth is independent of driving frequency, but increases with increasing temperature. The moderate temperature dependence is inconsistent with expectations from electron-phonon coupling alone, but consistent with charge noise arising in the device. The extracted level of charge noise is comparable with that expected from previous measurements of a valley-spin qubit, where it was hypothesized to be the main cause of qubit decoherence. Our results suggest a possible route towards improved valley-spin qubits.

cond-mat.mes-hall↗