Next-generation high-performance computing systems require high-bandwidth serial links to transport high-speed data streams among computational blocks. Optical links have recently attracted attention due to their low channel loss at high frequencies, requiring simpler equalization circuits than electrical links. The energy-efficiency of optical links can thus be significantly improved [1][2][3][4][5]. Broadband techniques such as inductive peaking are commonly used in highspeed optical transceivers for bandwidth enhancement at the expense of the chip area. Inductor-less receivers have been proposed [4,6] to reduce chip area but they usually consume more power or have lower data rates at given technology nodes.In this paper, we present two optical receivers that each consists of a pseudodifferential CMOS push-pull transimpedance amplifier (TIA), a DC offsetcancellation circuit, a limiting amplifier (LA) with interleaving active-feedback [6], and a T-Coil f T -doubler output buffer. The block diagram and experimental setup are shown in Fig. 8.4.1. The capacitance of the off-chip GaAs PIN photodetector (PD), which is wire-bonded to the CMOS receiver, is 100fF with 0.4A/W responsivity. The two optical receivers have identical designs except for the LA, in which two different inductive peaking techniques, conventional and sharedinductor, are designed and fabricated on the same die in 28nm CMOS technology. Figure 8.4.2 shows the circuit schematics of the pseudo-differential CMOS pushpull TIA with series-peaking inductors. The CMOS push-pull TIA has good signal gain and input-referred noise at low supply voltage because of current re-use of NMOS and PMOS [4]. Compared with CMOS inverter TIA in [4], the presented TIA employs a current tail to make the g m of M1~M4 refer to the bias current instead of the supply voltage for better supply noise rejection. In order to provide better single-ended to differential conversion for LA input, we include the crosscoupled pair M 7 and M 8 , which act as common-source amplifiers to provide negative voltage gain through the feedback resistor R F . The pseudo-differential configuration of the TIA provides better supply-noise rejection as well as jitter performance than the singled-ended TIA. The simulated gain of TIA is 46dBΩ and the input-referred noise is 2.5μA rms with 20GHz BW. The DC-offset cancellation circuit in Fig. 8.4.2 receives pseudo-differential outputs from the TIA and adjusts the output DC levels for the LA, based on the offset voltage provided by the LPF as shown in Fig. 8.4.1.Circuit schematics of LAs using both conventional and shared-inductor peaking are shown in Fig. 8.4.2. For the LA using conventional inductive peaking, two inductors L 1 are required for each stage. On the other hand, the LA using sharedinductor peaking requires only two inductors L 2 for every two adjacent stages, wherein the inductance value of L 2 is half of L 1 . The reason that L 2 can be only half of L 1 is because by sharing the inductor L 2 between two adjacent stages, the in-phase current...

No abstract

No abstract

Highly metastable pseudomorphic Ge0.3Si0.7 layers 570 nm thick were grown on Si(100) at ∼300 °C by molecular-beam epitaxy. The relief of strain in such metastable layers upon ex situ thermal annealing in vacuum is investigated by double-crystal x-ray diffractometry and MeV 4He channeling spectrometry. Upon isochronal annealing of 30 min, the strain relieves sharply at (375±25) °C, and reaches the thermal equilibrium value above 400 °C. Under isothermal annealing between 300 and 400 °C, the time evolution of the strain relief has the characteristics of a nucleation and growth transformation. The strain relief is very slow initially, increases approximately linearly as the strain is partially relieved, and saturates upon approaching equilibrium strain state. Two important results are drawn from the experimental data. First, a deformation-mechanism map is constructed from which the strain relief rate of a metastable GeSi/Si can be extrapolated for given stress state and temperature. Second, the rate of the strain relief when the strain is partially relieved increases with rising temperature, and follows an Arrhenius behavior as a function of the inverse temperature with a slope of 2.1±0.2 eV. This value coincides with the activation energy for dislocation glide in Ge0.3Si0.7. Furthermore, the strain-relief equation of a plastic flow model is solved and fits well the experimental strain-time dependence. One of the two fitting parameters, the time constant, has an Arrhenius temperature dependence. The slope, 1.9±0.2 eV, is assumed to be the activation energy for dislocation motion, and agrees with the previous value extracted from the simple rate-temperature dependence. In addition, as the strain is relieved, the x-ray-diffraction peak from the layer broadens and the channeling yield increases, confirming that the generation of misfit dislocations associated with the strain relief is accompanied by the generation of threading dislocations in the layer.

No abstract