Design And Reuse

Tuesday, July 15, 2008

Low power design for analog/mixed signal IP

Power reduction and management techniques using multiple clock and power domains, dynamic voltage and frequency scaling and power gating are effective for digital circuits but for analog design, lowering power consumption must be considered early in the design phase. Starting with the techniques for lowering the power consumption in analog circuits such as operational amplifiers, this article will then focus on low power design for high-speed serial interconnects. Different architectures for output drivers and methods such as level shifting, for ac-coupled systems such as PCIe, Serial ATA and XAUI will be discussed.

System designers are influencing the specifications of high speed serial interconnects and a good example of this can be seen with the emerging standards for the USB protocol: LPM and HSIC. USB is prevalent as the high-speed serial interconnect in portable devices such as smart phones and mobile Internet devices. The goal of link power management (LPM) is to reduce power consumption of USB devices and hosts, potentially extending battery life by at least 20 percent. HSIC or "high speed inter-chip USB" allows low power high-speed data transfers (480Mbit/s) using a source synchronous clocked serial interface. Both will be reviewed in this article.

Back to basics: The op amp

In the design of analog/mixed-signal IP many factors contribute to the overall consumption of power. The common methods include:

Simplifying the complexity of the circuit and using folded designs exploiting the complementary properties of NMOS and PMOS devices.

Taking conventional architectures and converting them into designs that consume less power with adaptive biasing.

Gearing the integrated circuit technology towards low power performance by using high Vt processes for example 65nm LP or 40nm LP. Although this may not necessarily reduce power in the active mode as more current is needed to drive the high-speed transmitter in the slower, low power technologies.

Decreasing transistor dimensions together with lowering the supply voltage.

Before delving into power reduction techniques for high speed serial interfaces, consider the case of an operational amplifier as the techniques applied here are pertinent to many other circuit examples.

1. The power consumption of the operational amplifier can be reduced by use of an architecture with only a single (differential) stage. This will reduce the current consumption of the device. However, a method of maximizing the gain, while preserving an acceptable bandwidth and slew rate are now required in the single gain stage.

2. The output stage could be designed to provide sufficient output drive while quiescently consuming as little power as possible.

3. Optimizing the biasing circuit will reduce the power consumption in the op-amp. This is achieved by reducing the internal stage currents by programming an external current in the form of a resistor outside the integrated circuit. Speed, voltage noise and junction leakage will now become major considerations for the designer as these parameters are affected by the value of the bias current programmed.

4. Two important factors that determine the maximum power dissipation in an integrated circuit are the technology used for the design and the type of application. A particular application for CMOS op-amps could be low power switched capacitor filters. If a lower power/low leakage CMOS technology such as 65LP or 40LP is used, then there are two important requirements in the op-amp design. First there must be enough current to charge the compensation capacitor and load capacitor in the required time. Second there must be enough current in the second gain stage transistor to maintain a phase margin of 45 to avoid ringing and degradation of the settling time. If the output current of this circuit is less than the quiescent bias current then this is known as a Class A circuit.

5. Quiescent power dissipation can be reduced by replacing Class A op-amps with Class AB and dynamic op-amps. The Class AB output stage is designed to be biased at small currents so quiescent power dissipation is correspondingly lower.

6. The basic two-stage differential input op-amp can be designed in the sub-threshold current region to minimize the current consumption.

Scaling issues: Is analog anti-Moore?

Whereas scaling of digital circuits is has been investigated in detail, the application of scaling analog circuits is still not that common. For example typical transistor dimensions in an analog circuit are a few multiples larger than 40nm minimum channels.

The sub-threshold characteristics of devices with channel lengths below 2m are very different to devices with larger dimensions. It has been observed that the current becomes exponentially dependent on drain voltage independent of VDS. This effect is sometimes referred to as "drain induced barrier lowering" (DIBL). If this effect can be avoided then IDdecreases exponentially as VGS is reduced below VT.

Scaling devices, and reducing the supply voltage accordingly, will not degrade open circuit voltage gain. Scaling dimensions but keeping supply voltage constant will, however, decrease the gain. The dynamic range of circuits such as op-amps fall because the analog signal range becomes limited due to the reduction in the power supply voltage. The problems of scaling down include fabrication challenges, limitations in the design of devices and circuits and the efficiency and distribution of power supplies.

Thermal noise will remain constant because the device transconductance remains constant under constant field scaling. The 1/f noise intensifies, but the effect of this can be reduced by translating the signal to a higher portion of the frequency spectrum, using chopper stabilization.

Low power guidelines

A set of guidelines can be developed for low power analog design, considering the op-amp as an example: the biasing circuitry, input, output and compensation stages can be examined.

The DC biasing circuitry for the op-amp must provide accurately determined and suitably regulated quiescent biasing currents at very low current levels. The current must be insensitive to changes in temperature, supply voltage and process tolerances.

The configuration of the input stage will dictate whether the op-amp can be used in a single low voltage (1.2V, or lower) supply application. Therefore the following properties are desirable (but not always possible) on a particular low power analog design:

Very low power supply operation using core devices Class B or AB output stage " this reduces the quiescent power dissipation particularly in a leaky process A rail to rail common mode input range A load-capacitance aware compensation scheme Capability to drive a small output load Bandwidth and slew rate commensurate with supply current Getting high precision by keeping offsets low, high input impedance, high CMRR and high PSRR.


Low power design example
The following is an example of a low-power, high-speed driver used in a USB 2.0 PHY. Power is kept very low with the use of a voltage-mode driver, shown in Figure 1.



In fact, for AC-coupled systems such as PCIe, Serial ATA and XAUI, a voltage-mode driver draws 25 percent of the supply current of a current-mode implementation. For example, a classical USB 2.0 driver requires nominal amplitude of 400mV for high-speed data transmission into a 45 load. The static power dissipation from the supply for a current-mode driver is approximately 60mW. Whereas, the power dissipation from the supply for a voltage-mode driver is approximately 30mW.

Another power-reduction technique is to optimize by level shifting down into the low-voltage core device domain as early as possible in the signal path so that device characteristics are leveraged as much as possible in order to reduce power. This strategy enables most high-frequency analog signal processing to be done in the low-voltage domain (for example in the USB, high-speed squelch detection and high speed receive function are implemented using low voltage core devices).

Low-power design is also important in sleep modes (non-functional modes) especially for mobile devices. For example, in the USB protocol the PHY can be held in suspend mode for long periods of time. As a result the power dissipation in this mode can become a significant portion of the total and must be scaled aggressively.

Power gating in digital circuits is often used where more aggressive power reduction is necessary. Power gating helps reduce both channel and gate leakage by collapsing the supply and eliminating the leakage current path. This can be implemented on a sub-block level by implementing collapsible, thick-oxide regulators that can be disabled in low-power modes.

Digital supply leakage often dominates the power down current and so many SoC's collapse the digital (core) supply rail in these modes as well. Doing so can cause serious side effects as analog level-shifters found in MSIP can often cause excessive current draw. Special care in the design of level shifters found in MSIP is necessary to avoid this. One example includes circuitry that detects when the digital power supply is collapsed; the circuitry then presets the analog control signals to the IDDQ state thereby eliminating unwanted leakage current.

Emerging low power standards for USB

For portable devices system designers are looking at ways to reduce power and one area where power can be saved is using low power versions of USB. Two standards are emerging: "Link Power Management" (LPM) and "High Speed Inter-chip USB" (HSIC).

LPM defines a new power sleep state between enable and suspend thereby conserving more power than the present suspend/resume mode. Also sleep mode occurs more often and faster reducing the transitional power states by three orders of magnitude. This reduces power consumption of USB devices and hosts and potentially extending battery life by at least 20 percent of the portable device.

HSIC is USB without the cable or the connector. It allows low power high-speed data transfers (480Mbit/s) using a source synchronous clocked serial interface. Low power is achieved with 1.2V LVCMOS signaling levels that is there no 3.3V signaling.

Summary

Novel circuit techniques can be used to reduce power significantly—enabling integration in very low-power mobile applications. Power reduction techniques include investigating the following:

1. Transmitter architectures 2. Analog signal processing in low-voltage domain 3. Sleep mode power reduction

Technology choice can impact power consumption although an LP technology may not necessarily give the lowest power for high speed serial interconnects. From a systems perspective applications that need to support low power embedded designs and portable devices such as smart phones and mobile internet devices will require new standards such as LPM and HSIC.

References:

"Building High-Quality, Mixed-Signal IP in 65-nm and Beyond" by Chong, Lam, Nandra, Toffolon, IP07, December 2007.

About the author

Navraj Nandra is director of product marketing for the mixed-signal products that include SERDES, USB and DDR2. Navraj holds a masters degree in Microelectronics, majoring in analog IC design, from Brunel University and a post-graduate diploma in Process Technology from Middlesex University.

No comments:

  

EE Times Semiconductor News

Design And Reuse - Industry Articles

EDA Blog

Technology Insider