基于高纯锗探测器的大质量弱相互作用重粒子（WIMPs）直接探测和无中微子双贝塔衰变（0νββ）观测是目前的研究热点。为了提高实验灵敏度，二者需要低本底条件、大质量探测器以及长期观测。中国暗物质实验（CDEX）目前正在中国锦屏地下实验室部署百公斤级高纯锗探测器阵列，将同时观测WIMPs和0νββ。本文首先为未来CDEX高纯锗阵列探测器设计了波形采样和读出电子学。相对于以往的采样率100 MSPS、采样精度14-bit的V1724商业电子学插件，本文采用了100 MSPS 16-bit和1 GSPS 14-bit 两种模数转化器（ADC）组成快慢系统。采用了低噪声ADC模拟前端驱动电路和低时钟抖动设计，其中100 MSPS ADC的有效位数（ENOB）为12.1-bit，优于以往的11.3-bit，并满足高纯锗探测器能量分辨的需要。1 GSPS高速ADC的ENOB为10.4-bit，满足高纯锗不同位置沉积能量事例的波形甄别需要。该系统目前已交付CDEX使用，并结合CDEX HPGe探测器检验了电子学系统的稳定性，为未来百公斤级和吨量级的实验奠定了基础。能量分析阈值是高纯锗WIMPs直接探测的重要指标。降低能量阈值有利于提高暗物质探测的灵敏度，其需要在一定的噪声误触发计数率下，提高物理事例的触发效率，即需要提高物理信号与噪声的甄别能力。由于以往基于模拟成形的幅度过阈触发方案，随着触发阈值的降低，不能有效地甄别物理信号和噪声，本文将基于神经网络的非线性甄别方法应用于物理信号与噪声的甄别。基于宽能高纯锗（BEGe）探测器实验平台，建立了误差反向传播的全连接神经网络，通过Monte Carlo波形模拟得到带标签的训练样本，利用训练好的模型参数，对实际探测器噪声和脉冲发生器（Pulser）产生的近阈值信号进行甄别，其效果优于以往的CR-RC线性甄别方法，可将能量阈值从600 eV降低至500 eV。该方法可推广应用于未来CDEX暗物质直接探测的在线触发方案中。能量分辨率是高纯锗探测器0νββ观测的重要指标，提高能量分辨率有利于提升0νββ观测的灵敏度。结合CDEX高纯锗探测器，在以往“模拟成形+波形数字化”的方案下，使用了本文研制的波形采样电子学，因其较好的ENOB指标，能量分辨率初步测试结果优于以往的V1724商业电子学插件，由5.5 keV @2614 keV提升至4.0 keV。进一步地，采用了“直接波形数字化+线性最优滤波”方案，实现了能量分辨率由4.0 keV至3.8 keV的再次提升，为未来CDEX 0νββ实验观测的波形数字化和成形方案提供了参考。

Direct detection of Weakly Interacting Massive Particles (WIMPs) and observation of neutrinoless double-beta decay (0νββ) based on HPGe (High-purity Germanium) detectors are current research hotspots. To improve the experiment sensitivity, low-radiation background, large-mass detectors, and long-term observations are required. The China Dark Matter Experiment (CDEX) is now deploying 100 kg HPGe detector arrays at the China Jinping Underground Laboratory, where both WIMPs and 0νββ will be observed.First, this paper designs the waveform sampling and readout electronics for the future CDEX HPGe detector arrays. Compared with the previous V1724 commercial electronics with 100 MSPS sampling rates and 14-bit resolution, this article uses 100 MSPS 16-bit and 1 GSPS 14-bit analog-to-digital converters (ADC) to form a fast-slow system. An ADC front-end driver with low noise and low clock jitter design is adopted. The effective number of bits (ENOB) of the 100 MSPS ADC is 12.1-bit, which is better than the previous 11.3-bit and meets the requirement of HPGe detectors’ energy resolution. The ENOB of the 1 GSPS high-speed ADC is 10.4-bit, which meets the requirement of waveform discrimination of energy deposition at different locations in the HPGe detector. The system has been provided for the CDEX and the stability has been verified with the CDEX HPGe detector. It lays a foundation for the future CDEX with 100 kg and ton-scale HPGe detectors arrays.The energy analysis threshold is an important index for the direct detection of WIMPs. A lower energy threshold can improve the sensitivity of dark matter detection. Under a fixed noise false trigger count rate, improving the triggering efficiency of physical signals is in demand. Therefore, it is essential to improve the ability to discriminate between physical signals and noise. The previous trigger system is based on an over-threshold of analog shape amplitude. As the trigger threshold decreases, it cannot effectively distinguish between physical signals and noise. This paper applies a nonlinear discrimination method using a neural network. Based on a Broad Energy Germanium (BEGe) detector, a back-propagation and fully connected neural network is established. Labeled training samples are obtained through Monte Carlo waveform simulation. The trained model parameters are used to determine near-threshold waveforms generated by Pulser and baseline noise from the actual detector. Its discrimination performance is better than the previous CR-RC linear discrimination method and the energy threshold can decrease from 600 eV to 500 eV. The nonlinear determination method can be applied in the online trigger scheme of the future CDEX.The energy resolution is an important indicator for the 0νββ observation of HPGe detectors. Optimizing the energy resolution is effective to improve the sensitivity of 0νββ observation. Combined with the CDEX HPGe detector, under the previous "analog shaping + waveform digitization" scheme, the waveform sampling electronics developed in this paper are deployed. Due to its better ENOB, the preliminary energy resolution is better than the previous V1724 commercial electronics, from 5.5 keV @2614 keV to 4.0 keV. Furthermore, the "direct waveform digitization + linear optimal filtering" is proposed, which optimizes the energy resolution from 4.0 keV to 3.8 keV. It provided a reference for the waveform digitization and shaping in the future CDEX 0νββ observation.