TORONTO -- Reducing power consumption in memory is an ongoing challenge, but designing more energy-conscious controllers is also critical depending on the application.
Silicon Motion recently announced its SM2246EN SATA (6Gb/s) client SSD controller now supports Micron’s 16nm 128Gb MLC NAND flash. In addition to highlighting its performance, the company also emphasized its ultra-low power consumption, which on average is 56mW.
It’s the application that determines whether the power consumption in an SSD controller is a critical system feature, said Jim Handy, principal analyst at Objective Analysis. “The power consumption for high-end enterprise SSD controllers is something that nobody really cares about.” When it comes to SSDs for client applications such as PCs, it’s more important to address power use to extend battery life since a lot of SSDs are getting used in ultrabooks and Macbook Air computers.
“It’s not just the power the controller expends,” said Handy, “It’s how the controller manages the memory.” The power that the SSD uses, he said, is very much influenced by how much the NAND flash is written to and erased because those processes are the high power aspect of a NAND chip’s operation. If you reduce the writes of the SSD, the performance is more consistent and improves endurance.
“The SSD controller makers focus intense efforts on minimizing the amount of write traffic they have in the SSD, and that in turn minimizes the power,” said Handy. Controllers have their own power requirements that are heavily affected by their microprocessors, and a balance must be struck. “Are you going to the make the algorithms more sophisticated at the expense of burning up more power in the controller, or are you going to make the algorithms less sophisticated and burn up more power in the NAND flash chip?”
Robert Fan, VP and general manager of Silicon Motion USA, said lowering power consumption of controllers has become more critical as more SSDs are being designed to go into smaller devices. The company has made it a priority in its design process to architect the controller to consume less power. “The idea is to minimize consumption in the entire subsystem.”
Silicon Motion uses a number of techniques to that end. One is clock gating, which shuts down circuitry when it’s not in use, as well as using algorithms and direct memory access to transfer data without going through the CPU, which Fan said consumes the most power. Using four channels instead of eight also helps reduce power consumption, he said.
In enterprise systems, performance takes higher priority over power consumption, which is a “nice to have” feature, said Fan, but in portable devices power use is more critical.
For Marvell, power consumption is a consideration in the design of all its controllers, said Zining Wu, Marvell’s CTO, starting at the architecture level. “We pay attention to this from the beginning.” Power consumption of controllers is getting more attention due to the growth of embedded systems, wearables and the Internet of Things. Earlier this month, Marvell announced a new wireless product line for IoT applications. It includes the company's MW300 WiFi microcontroller. In May, it introduced its 88SS1074 SATA SSD controller for use with 15nm triple-level cell (TLC) NAND flash in client and enterprise SSDs; like Silicon Motion, Marvell is also emphasizing lower power consumption, and Wu said algorithms play a big role in addressing power usage.
Ultimately, Wu said, careful design of algorithms and architectures can allow for reduced power use throughout the entire system. There are limitations and boundaries, he said. “Each process has its pros and cons.” Smaller geometries do save power, but standby power tends to go up, “so you do have a tradeoff.”
With devices running more applications, Zu said, more memory is going to be required, which offsets the gains made by CMOS processes. “We do get better and better in our designs, and with each generation we can improve architectures and algorithms to save power.”