Issue link: https://iconnect007.uberflip.com/i/666328
40 The PCB Design Magazine • April 2016 is the most straightforward approach. It doesn't matter which topology you use, though—both fly-by and double T-topologies should work fine. If you are using a DDP device, then dou- ble-T topology works better than fly-by in terms of delivering a better system margin. During a write cycle, using the fly-by topol- ogy, data strobe groups are launched at separate intervals to coincide with the clock arriving at memory components on the SODIMM or PCB, and must meet the timing parameter between the memory clock and DQS defined as tDQSS of ± 0.25 tCK. The PCB design process can be sim- plified using the leveling feature of the DDR3/4. The fly-by, daisy chain topology increases the complexity of the controller design to achieve Figure 3: Fly-by topology for clock/address/command routing. leveling but fortunately, greatly improves per- formance and eases board layout for DDR3/4 designs. It is not that you have to use fly-by write leveling, because it is a feature of DDR3 and DDR4, but rather that you have to use write leveling in order to allow fly-by routing. There is also no reason not to use the write leveling training for a T-topology in order to optimize the write strobe to clock timing. With this you can adjust slight differences in CA timings and avoid hard coding the skews that you normally have to manually take care of on the strobe to clock delay. Fly-by topology is similar to daisy chain or multi-drop topology, but has very short stubs, to each memory device in the chain, to reduce the reflections. The double T-topology was used for DDR2 and had a downside in that the impedance discontinuities, due to branching along the traces, caused obvious margin losses. T-topology also tends to have overshoots, while the levels for fly-by are terminated and there- fore do not reach the full swing voltage rails. Also, the length of the stubs has an effect on the maximum bandwidth of the transmission line. If you are employing high-frequency DDR4, then the bandwidth of the channel needs to be DDr3/4 fLy-By vs. t-topoLogy routing