• Tidak ada hasil yang ditemukan

Chapter 1 Introduction

1.4 Countering Variations

1.4.1 Review of Existing Techniques

The commonly used approach to solve the issue of performance degradation due to variations is to design more variation tolerant systems. This entails adopting architectures and circuit topologies, which are less sensitive to process and mismatch variations. These techniques have been widely adopted in CMOS analog and digital designs over the years.

[19] discusses a new technique where there exists a gate bias voltage for which the vari- ations in carrier mobility compensates the (VGS-Vth) variation when temperature fluctuates.

At this optimum voltage, the drain current can become less dependent on temperature. In fact, to reduce variations for some processes, the supply voltage needs to be lowered from

nominal for optimum operation. In [20] a low frequency CMOS oscillator has been reported where a bias generator circuit generates a process dependent voltage, which is then utilized to control the VCO through a temperature compensating block. In [21], a robust CMOS operational amplifier is designed where a constant gm stage is implemented without the necessity of matching between n and p-channel transistors. A tuning curve linearization technique and switched capacitor current source were used in [22] to reduce sensitivity of a GPS VCO to PVT variations. Adaptive body biasing was demonstrated in [23] to reduce leakage power in nm domain CMOS circuits.

These techniques discussed above are difficult if not impossible to implement at RF/mm- wave frequencies. Efforts have been made over the past several years to incorporate tech- niques like tunable matching network and adaptive biasing among others to improve tunabil- ity of RF transmitting/receiving frontends. In [24], a CMOS VCO has been demonstrated to be less sensitive to process corners using an iterative design technique. [25] proposes a tunable output and input matching network based LNA at 2.4GHz which can be used to compensate for optimum performance across process corners. The technique leads to sig- nificant improvement in variation against process and mismatch, as shown in Figure 1.10.

A PVT tolerant LC-VCO was reported in [26] where a process dependent reference voltage

Figure 1.10: Reduction in standard deviation of LNA gain as reported in [25].

is used to control the supply regulator of the VCO, thereby reducing the sensitivity of the

to reduce PVT variations, their application is limited in high frequency RF/mm-wave CMOS design, where some of these techniques tend to severely affect the performance of the system.

Moreover, the systems discussed above tend to be insensitive to only a few types of variations and often cannot counter dynamic or unexpected variations like aging and/or environmental variations. Thus, to truly build high-frequency systems that are tolerant against a wide variety of variations, both static and dynamic, we need a completely autonomous closed loop system. We discuss some of the salient aspects of this new approach in the following sub-section.

1.4.2 Self-healing to Counter Variations

A more scalable approach to deal with performance hits due to variations is to sense the degradation caused by variations and then change the system performance by using various knobs. The ability to dynamically sense and actuate critical high-frequency blocks of the sys- tem eliminates the additional design complexity of variation insensitive circuits/techniques.

We term this closed-loop technique as self-healing. Figure 1.11 shows the basic block diagram of a representative RF/mm-wave self-healing system where data from sensors are processed using a digital processing block which then decides the optimum actuation settings.

Inputs Outputs

Figure 1.11: Conceptual self-healing mm-wave system.

Due to the digital driven aggressive scaling of CMOS in the past few years, most modern CMOS processes offer very limited high-frequency analog modeling of transistors. Iterative design based on measurements from such test transistors significantly slows down the design cycle and increases production costs. In addition, as discussed in the previous sections, the variations increase with further process scaling. An ideal closed-loop autonomous self-healing system should address both these problems, as shown in Figure 1.12, thereby improving overall yield.

Figure 1.12: Self-healing (a) improving performance as well as reducing variations and (b) improving yield.

The self-healing loop involves integrated sensors that detect the performance of the mm- wave circuit. Several stringent requirements apply to these sensors, especially since they are subject to the same set of variations as the mm-wave circuit. Robustness of these sensors is es- pecially critical, since their outputs should be true representations of the circuit performance and not dependent on variations within the sensor. To harness the vast digital processing power of modern CMOS processes, the self-healing algorithm must be implemented digitally.

The analog output voltages of these sensors thus need to be converted to digital form using on-chip analog-to-digital converters (ADC) that send data to an integrated digital core. The

self-healing core then controls the RF system through digital to analog converters (DAC) that set the actuation points of the circuit. The actuators also need to cover a wide enough actuation space to be able to heal against unforeseen variations in the chip performance.

Moreover, these actuators and sensors need to have minimum impact on the high-frequency performance of the system. This loop may be iterated until an optimum actuation state is reached.

Digital systems have deployed similar concepts primarily because sensors and actuators in the form of digital control can be easily integrated into the system without significant loss in performance. [27] describes a system which utilizes several adaptive processing units in addition to regular ones and periodically turns off the system and performs a functional verification. If necessary, the regular processing units are replaced with their adaptive coun- terparts to enable optimum system operation. The LNA in [28] incorporates a tunable matching network and a input return loss sensing mechanism in the form of a small valued resistor in the source of the input transistor to enable a closed loop system of input match auto-calibration. A reconfigurable test scheme is reported in [29], where the LNA perfor- mance is sensed by re-structuring it in a feedback configuration to produce an oscillation signal, which is then utilized as test signal to test the amplifier. Bias control and match- ing network tuning is used subsequently to heal the LNA back to its desired performance.

As shown in Figure 1.13, the yield of a SiGe LNA was improved from 87% to 97% after self-healing using this technique.

A self-healing mm-wave power amplifier is reported in [30] where, based on output power levels, the PA bias is adjusted to provide constant gain. An extension of the work in [31] uses the adaptive biasing scheme to improve P-1dB of the 60GHz PA by 5.5dB from nominal.

On-chip PVT compensation has been demonstrated in a 2.4GHz LNA [32], where the input of the LNA is switched between an off-chip and on-chip VCO during measurement and calibration phases, respectively.

Figure 1.13: Self-healing to improve yield of a SiGe LNA [29].

Dokumen terkait