Powering Ethernet, Part 1: Designing for low power consumption in operation
Analyzing power consumption in Ethernet circuitry shows it to be far from efficient. This is in part because Ethernet consumes similar energy during both traffic and idle periods, and in fact, idle periods typically account for more than 97 percent of the time. This was key to determining where improvements could be made to reduce power consumption, when investigated by an IEEE task force; resulting in the standardization of IEEE 802.3az, or Energy Efficient Ethernet.
Energy Efficient Ethernet shows great promise to universally succeed where earlier attempts to reduce idle period power have somewhat failed with methods such as Wake-on- LAN. Complimenting Energy Efficient Ethernet, additional power savings can also be made both during normal traffic and link down. This paper outlines where the current is consumed and how to design for the lowest power consumption, both in operation (Part 1) and standby (Part 2), since calculating the power consumption of an Ethernet circuit is not always straightforward.
The first step towards achieving low power Ethernet designs is understanding where the power is dissipated. In any Ethernet device, the major power dissipation is from the PHY transceiver (Figure 1).
Figure 1: Power dissipation from the PHY transceiver: power is often consumed both internally to the PHY and externally in the transformer
In the case where Ethernet datasheets publish the device only current consumption, calculating the total circuit current consumption requires the designer to add typically around 40mA per 100Base-TX or 70mA per 10Base-T PHY for dissipation in the transformers. As a result, a lower device only consumption at 10Base-T will rarely equate to lower total circuit current consumption, relative to 100Base-TX mode. Micrel’s latest PHY technology employs voltage mode techniques which not only reduces power consumption typically by 50% but dissipates all current within the PHY itself; resulting in the total power consumption = device power consumption (no dissipation externally in transformer).
A designer must consider the following two modes: Normal Operation and Standby, when trying to further reduce power consumption.
Ethernet Power Consumption (Normal Operation)
So, what is the definition of normal operation for an Ethernet network? Is it 100 percent utilization, 50 percent utilization, or 10 percent utilization? The reality is long quiet periods followed by relatively short bursts of traffic (Figure 2). During these quiet periods, Ethernet power consumption might be expected to significantly drop, however, this turns out to not necessarily be the case.
Figure 2: Ethernet Network Utilization – Typical
1000Base-TX and 100Base-TX are both designed so that the link partners are continually synchronized to each other. To enable this, when no traffic is being transmitted the PHY will automatically send out IDLE symbols (11111 5B code). As a consequence, during any quiet period the PHY transmitter is still operating in a manner similar to full traffic – meaning it consumes a similar amount of power.
It is strongly advisable with multi-port Ethernet devices, to disable any unused port (PHY), since simply by connecting to a link partner, around 40mA current is consumed even with no traffic present. The port can usually be disabled via the internal register map (Register 0h bit 11 of the IEEE Defined PHY Registers) and will typically save a further 15-20mA of device current.
10Base-T operation differs during quiet periods, since when no traffic is present, the PHY transmitter does not transmit out any IDLE symbols. Instead, it sends out a single link pulse approximately every 16ms, designed simply to keep the link alive. The power consumption of the PHY itself during a quiet period in 10Base-T operation will not reduce significantly, but the current consumed externally in the transformer will reduce to negligible, saving around 70mA per PHY compared to full traffic.
Cable Length, Driver Strength
For Ethernet conformity, the PHY transmitter must adhere to fitting within the defined limits of the mask (Figure 3).
Figure 3: 100Base-TX IEEE802.3 Output Eye Diagram
This waveform is designed to ensure that the PHY is capable of operating up to a minimum 100m of CAT5 grade cable. As a consequence, the PHY output drive strength is fixed at this criterion, consuming maximum power, independent of the actual length of cable connected. There was no provision to adaptively adjust the drive strength dependent on the cable length, making this a major obstacle in the original IEEE 802.3 specification with respect to energy saving.
However, in practice, many applications do not require the capability of 100m-cable reach and can guarantee a much shorter length. A simple change to the circuit can reduce the PHY transmitter current drive, typical set by a resistor, from the standard ±1V amplitude of the 100Base-TX signal down by up to 50 percent and still operate error free over a 10-20m reach (typical for automotive networks). For example, doubling the resistance will half the typical 40mA 100BT drive current to around 20mA per port. Longer cable reach could be achieved while operating at reduced current drive by installing higher quality cable e.g., CAT6 or above, that exhibits lower attenuation. System costs, however, are increased.
The use of cable diagnostics features, such as Micrel’s LinkMD®, offer has the ability to measure the connected cable length, using time domain reflectrometry techniques. This allows designers to intelligently adjust the drive strength according to the cable length, thereby improving the power consumption efficiency. An additional benefit of reducing the PHY current drive strength is the reduction in electromagnetic interference (EMI) radiated emissions.
Another area important to explore when ensuring maximum energy efficiency is the power management of the Ethernet device. Many modern devices operate using a single voltage, typically 3.3V, and provide internal regulation for core voltage(s). This provides the customer with a simpler implementation – but has a significant impact on power efficiency. If possible (often not unfortunately), disable the internal linear regulator and supply the lower core voltage externally (often already available on the board).
Micrel’s KSZ8863/73 3-Port Ethernet switch offers an example:
Table 1: KSZ8863/73 Typical Current Consumption
Using the internal regulator, total Power Consumption is 3.3V × 121mA = 400mW. But if one disables the internal regulator and operates using external 1.8V supply, then the total Power Consumption is 3.3V × 21mA + 1.8V × 100mA = 249mW. This improves power efficiency by an impressive 38 percent.
Sponsor video, mouseover for sound
Power consumption in electronic applications has increasingly been viewed as critical, with worldwide legislation forcing manufacturers to improve energy efficiency. As we have seen here, power consumption can be reduced during normal operation, but this is only once aspect of achieving Energy Efficient Ethernet. Not only is the power consumed when devices are in operation but also during ‘standby’ periods. Power savings during standby operation will be covered in Part 2 of this series.
About the author
Mike Jones has over 25 years of experience in high-tech design in the semiconductor industry. Jones is currently based in Newbury, U.K., where he is Product Marketing Director, responsible for Micrel’s Automotive and Industrial LAN Solutions. Prior to Micrel, Jones worked for several high tech companies, in various engineering roles. including as principal engineer at BT and Fijitsu Telecommunications where he gained more than a decade in design experience in SONET, SDH and PDH systems. Jones also held position of senior FAE and Product Marketing Manager with Micrel for fourteen years prior to my current position.
Jones graduated in 1990 with a 1st in class honors degree in Electric Systems Engineering at Aston University in Birmingham, UK.