Mode-locked fiber lasers output ultra-short pulse trains with extremely high temporal stability, showing great potential in systems that require precise timing synchronization, such as pump-probe experiments, high-speed analog-to-digital conversion, large-scale timing distribution and coherent combination. The fiber lasers are usually simpler, less costly, more efficient and more robust to the environment than solid state lasers, making them a better option for real-world applications. With the atto second temporal resolution of the balanced optical cross-correlation (BOC) method, timing jitter of mode-locked fiber lasers has been carefully measured and optimized over the last decade. However, due to the inherently large amplified spontaneous emission noise in the long gain fiber and broad pulse width inside the laser cavity, the quantum-noise-limited timing jitter of mode-locked fiber lasers is still much higher than that of the solid state lasers. In order to further optimize the timing synchronization of mode-locked fiber laser, larger locking bandwidth is required to suppress the low-frequency timing jitter, which contributes significantly to the total amount of residual timing jitter. In this work, tight timing synchronization between two mode-locked Yb-fiber lasers is achieved via a feedback loop built on an intra-cavity electro-optic phase modulator. Both lasers work in the stretched-pulse regime, which has been proven to support the lowest quantum-noise-limited timing jitter of mode-locked fiber laser. The output of the BOC system provides a timing error discriminator of 40 mV/fs, corresponding to 13 as resolution within the integration bandwidth. When the pulse trains from both lasers are successfully synchronized, the residual timing jitter can be measured with the same signal as that used for timing synchronization Based on the residual timing jitter measurement, the intra-cavity dynamics of the laser and the locking parameters of the feedback loop can be further optimized and a tight synchronization with 400 kHz locking bandwidth is finally achieved. When performing the integration from 1 Hz to 10 MHz, the residual timing error is as low as 109 as, corresponding to 77 as averaged timing jitter of each laser. A parallel out-of-loop single-arm cross-correlation measurement is also performed to test the validity of the in-loop results, and both measurements agree with each other.