Issue link: https://iconnect007.uberflip.com/i/886239
October 2017 • The PCB Design Magazine 49 The problem with DDR5 topologies tends to be ringing—not loss. Serial channels are point to point, and discontinuities are stubs are sharply minimized to maximize signal quality. DDR5 topologies, by comparison, cannot afford to manage impedance and discontinuities as tight- ly—it's just not economically feasible. Because DDR5 topologies are multi-drop, multi-DIMM, stubs are long and reflections are huge by com- parison. That's why we're so interested to see how the use of DFE plays out in DDR5, because DFE technologies are uniquely suited for deal- ing with significant reflections in the few bit times that precede the main signal. Shaughnessy: How would users use AMI models to determine the best equalization settings for their systems? Katz: There are two users, and they want to know two different things: The systems designer wants to know that for a given channel there is a best set of equalization settings that will opti- mize the DDR5 channel performance (smallest BER, largest time and voltage margins), and that the BER and timing and voltage margins meet the system performance requirements. They also want to know that whatever training algo- rithm the controller uses to optimize the chan- nel at hardware runtime will end up with the settings needed to properly equalize the chan- nel. So the systems designer wants to be able to predict the effect of the hardware training al- gorithm on system margin without necessarily modeling that training algorithm literally. But the device vendor wants to design and validate the hardware training algorithm and prove that it will reach or at least come close to the best performance of the channel. They do want to model the training algorithm literally, or very nearly so. Shaughnessy: It sounds like this whole thing has a lot of layers. Maybe we should talk again after DDR5 launches, to see how people are do- ing with this? Katz: This is an incredibly detailed and inter- esting problem. What the actual hardware does and what the simulation models do will not be the same. Effectively managing the details of those differences will be critical to maximize the design's chance for success. Shaughnessy: Thanks for your time, Walter. Katz: Thank you, Andy. PCBDESIGN SISOFT PREPARING FOR DDR5 SIMULATION NEXT YEAR Researchers have created a min- iature 'superhero' robot capable of transforming itself with different 'outfits' to perform a variety of tasks. Inspired by origami, scientists from the University of York and the Massachusetts Institute of Technolo- gy (MIT) have developed a magnet- controlled shape-shifting device which can walk, roll, sail on water or glide. Dubbed "Primer," the cube-shaped robot carries out these actions by wearing different exoskeletons— accessories which start out as sheets of plastic that fold into specific shapes when heated. After Primer finishes its task, it can shed its "skin" by immersing itself in water, which dissolves the exoskeleton. Professor Daniela Rus, Director of MIT CSAIL and Principal Investi - gator on the project, said: "If we want robots to help us do things, it's not very efficient to have a dif- ferent one for each task. With this metamorphosis-inspired approach, we can extend the capabilities of a single robot by giving it different accessories to use in different situations." While robots that can change their form or func- tion have been developed at larger sizes, generally it's been difficult to build such structures at much smaller scales. What is also new about Primer is its ability to switch between many, rather than several, different forms. Scientists Develop Shape-Shifting 'Superhero' Robot