Concepedia

Publication | Closed Access

18.5 A 54MHz Crystal Oscillator With 30× 18.5 Start-Up Time Reduction Using 2-Step Injection in 65nm CMOS

11

Citations

4

References

2019

Year

Abstract

The start-up time of crystal oscillators (T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> ) is a major bottleneck in reducing the average power of heavily duty-cycled wireless /wireline communication systems [1]. Among all the reported schemes to reduce T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> , techniques that increase initial noise amplitude by injecting a surge of energy into the crystal resonator are shown to be most effective [1-3]. These approaches are proven to be robust if the frequency of the injection signal is equal to the crystal oscillator (XO) frequency (F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ</sub> = F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X0</sub> ), which is difficult to achieve across PVT with on-chip oscillators. Any mismatch (ΔF = F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ</sub> - F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X0</sub> ) even as small as a few 100 ppm can greatly increase T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> . Sweeping the injection frequency using a chirp oscillator [2] or dithering the injection frequency between two values [1]can partially alleviate this issue but because ΔF ≠ 0, this only reduces T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> to about 14× the theoretical minimum in [2]. On the other hand, it was shown in [3] that the use of a precise injection period T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ,OPT</sub> can help reduce T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> even in the presence of large ΔF. T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ,OPT</sub> must be chosen such that current in the motional branch of the resonator, i <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</sub> (t), reaches its steady-state value, I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS</sub> (i <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</sub> (T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ, OPT</sub> ) = I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS</sub> ) as shown in Fig. 18.5.1 [3]. However, small T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> and large tolerance to ΔF can be achieved only when I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS</sub> is very small, which translates to small XO output amplitude (V <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">XO</sub> <; 200mV) and degraded phase noise. For example, as illustrated in Fig. 18.5.1, T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> ≈ T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ,OPT</sub> because i <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</sub> (T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ, OPT</sub> ) = I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS1</sub> even when ΔF is as large as 1000ppm. However, for I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS2</sub> > I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS1</sub> , no TINJ can ensure i <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</sub> (T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INJ</sub> ) = I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m,SS2</sub> if ΔF > 1000ppm, thus greatly increasing T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> . Therefore, ΔF must be small (<;500ppm) to achieve a large V <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">XO</sub> even with precisely-timed injection, a condition that is difficult to meet in practice even with the best-reported temperature-compensated on-chip oscillators [4]. In view of these drawbacks, we present a robust 2-step injection technique that can tolerate large ΔF and achieves close to theoretical minimum T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> , large output swing and excellent phase noise. Fabricated in a 65nm CMOS process, the prototype XO achieves T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> of less than 20 μs across the - 40°C to 85°C temperature range, which is within 1.5× of the theoretical minimum and represents an over-30× reduction in T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">START</sub> compared to that of a normal/uninjected XO.

References

YearCitations

Page 1