2016 · 14 citations · 4 references
EngineeringComputer EngineeringElectrical EngineeringAnalog ElectronicsMicroelectronicsVlsi DesignClock ManagementCmos Circuit DesignClock GenerationFrequency CompensationMultipliersPhase MarginPhase-locked LoopsPower-efficient ArchitecturePower-efficient DesignRf Cmos Design
Phase-locked loops (PLLs) are de-facto clock generators in analog, digital, RF, and embedded systems to generate a high frequency output clock from a low frequency reference clock. Modern systems-on-chip (SoCs) require many such PLLs that cater to multi-core processors, memories, IO interfaces and power management. A ring-oscillator-based analog charge-pump PLL offers a flexible and power-efficient way to implement such clock multipliers. However, frequency compensation of analog PLLs requires a large loop filter capacitor that occupies significant area. The area penalty is further exacerbated in deep sub-micron CMOS processes due to increasing oscillator gain and gate leakage. For example, a 2GHz PLL in 65nm CMOS requires a 90pF capacitor (R=1kΩ) to achieve 10MHz bandwidth and 70° phase margin. Assuming a capacitor density of 1fF/μm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , a 90pF capacitor occupies an area of 0.09mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . Digital PLLs (DPLLs) offer an attractive means to eliminate the loop-filter capacitor [1, 2]. However, quantization error added by its loop components degrades the performance in multiple ways. First, it makes the DPLL loop inherently non-linear causing the steady state to be a bounded limit cycle, which manifests as deterministic jitter (DJ) at the PLL output. Second, it introduces conflicting noise bandwidth requirements, which makes it difficult to achieve low jitter in a power efficient manner. For instance, suppressing TDC quantization error by reducing PLL bandwidth increases the DCO phase noise contribution. As a result, DCO power must be increased to lower its noise contribution. This is especially problematic in ring-oscillator-based DPLLs. Finally, a high-resolution digital-to-analog converter (DAC) needed to interface the DLF output to the ring oscillator typically occupies large area and negates some of the area benefits of DPLLs [1, 2]. In view of these drawbacks, we present a PLL architecture that combines the advantages of analog (no quantization error) and digital PLLs (small area and scalability) by using a time-based integral path. The prototype PLL requires neither a high resolution DAC nor a capacitor and fits in 40μm×52μm (0.0021mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) active area. It achieves 3.8ps <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">rms</sub> integrated jitter at 2.2GHz and consumes 1.82mW power.
4