Regarding Ethernet or 10GBASE-T, from 2017 [10GBASE-T is finally popular?], We delivered a total of 11 times and 2 extra editions.However, Ethernet for copper wiring by twisted pair is up to 10GBASE-T, and 25 / 40GBASE-T has not yet been put into practical use.Let's continue with the content of the May 2021 meeting of the 802.3 Beyond 400 Gb / s Ethernet Study Group.Last time, I wrote that there weren't many suggestions for Optic, but Huawei's Maxim Kuschnerov and Lin Youxi gave a comparison of PAM4 and PAM6 in a 200G lane.Regarding PAM4 and PAM6, "200G x 8 1.6Tbps, x4 800Gbps transfer will be realized in 2023?In Google's presentation at the March meeting introduced in 』, the comparison was as follows.What was presented at the May meeting is the result of a more detailed simulation of this.First, from the simulation around the BER on the receiving side.Below is a graph of the results of applying a 4th order Bessel Filter and a 2nd order Gaussian Filter.Compared to PAM6, PAM4 has a sensitivity (sensitivity, or received power) of about 2 dB with a Bessel Filter and about 1.5 dB with a Gaussian Filter.In other words, if the BER is the same, the advantage of PAM4 is that the received power can be reduced by 1.5 to 2 dB.Next is the sender.In the 200G generation, two types of light emitting elements, EML and MZM, are promising as mentioned at the meeting in March mentioned earlier, but BER when PAM4 / PAM6 is mounted on each light emitting element. The following slides evaluated.Of these, the BER in MZM on the left is not much different, but PAM4 is a little better.However, as mentioned in the footnote, it tends to be slightly non-linear in comparison with EML, and PAM6 is more susceptible to this tendency.On the other hand, when EML is used (hereinafter referred to as the right), the BER divergence is quite large.With PAM4, the BER decreases as the ROP increases, and with B2B it drops to around 5.0E-5, and even at 5km it drops to around 1.0E-4, while with PAM6 it stays around 5.0E-3.Moreover, as mentioned in the footnote, if the development of components progresses and the bandwidth of DAC and EML expands, the difference in BER between PAM4 and PAM6 will rather widen.Next is the "MPI (Multipath Interference) penalty", which is a loss comparison when multiple Links are connected with a patch code, as shown in the figure on the upper right of the slide below.The mystery remains around the fact that the trend of the graph cannot be understood (the vertical axis is the sensitivity penalty, so the smaller the MPI penalty, the smaller the sensitivity penalty, but the graph is the opposite), but according to Kuschnerov et al., PAM4 Then, the penalty in Triple link is within the allowable range (in short, it depends on the loss of patch code), but in PAM6, Triple link is impossible because the MPI loss becomes too large.The slide on the right below compares the characteristics of CD (Chromatic Dispersion) when using PAM4 at 224 Gbps.At wavelengths of 1270 nm or 1330 nm, the variance of the characteristics becomes a little too large, and the content is that the 5th-order FFE is not enough for correction.Conversely, if the wavelength is 1300 to 1324 nm, the characteristic value is expected to fall within the dispersion range defined by ITU-T (“Table 1” on the left).By the way, two digressions from here.The first is that Google has made a request for a "200Gbps Passive Copper Cable."The motive is that "DAC (Direct Attach Copper), that is, Ethernet with copper wiring, is widely used in the Edge Aggregation Block and inside the rack already in the data center.To achieve this, by renewing the cable itself, PCB (wiring on the printed circuit board), connector, VIA (through terminal) and package, wiring inside the package, etc., if the total loss is 43 GHz, it is about 36 dB. , It is said that it will be about 50 dB at 53 GHz.By the way, in order to realize 200G (actually, it is over 200G because it includes FEC), if it is PAM4, it is 53G x 4 (this is a provisional value, so it may be even higher such as 56G depending on the result of examination). However, with PAM6, it can be realized with 43G x 6.Given this 36dB loss, Google argues that PAM6, which is quite realistic on the contrary, should be considered a little more.Another digression is the trend of OSFP, which was mentioned earlier in "Trends of OSFP MSA and Two IEEEs Related to 800G Ethernet".At least with the current Rev 4.0, although it can cover up to 800G, there is no touch on 1.6T, and the Signal Rate was supposed to be up to 100Gbps.However, the specification work is still in progress, and it seems that a new connector standard called "OSFP-XD (eXtra Density)" is planned.Actually, where I heard this is Andy Bechtolsheim (Arista Networks founder and Chief Development Officer, Sun Microsystems founder and Chief System Architect) who was held on the second day of "Hot Interconnects 28". This is the keynote speech.The OSFP-XD is an ambitious specification that allows signals from 16 lanes to be used in the first place, and also takes into consideration up to 200 Gbps per lane, that is, 1.6T and even 3.2T beyond that.Actually, I didn't notice that there was no release from OSFP, but it seems that it was first released at the "OFC 2021" workshop held in June 2021.Free technical writer.We have a wide range of fields of expertise, from CPUs, memories, and chipsets to communications, operating systems, databases, and medical care.The homepage is http://www.yusuke-ohara.com/Facebook and Microsoft DC operators preceded 400GbE, Beyond 400G Study Group insists on standardization of 1.6T at the same time as 800G1.6Tb / s Ethernet specifications to be formulated at the same time as 800Gb / s 200Gb / s lane product shipment around 2027?Facebook and Microsoft DC operators preceded 400GbE, Beyond 400G Study Group insists on standardization of 1.6T at the same time as 800GTrends in OSFP MSA and two IEEEs related to 800G Ethernet"400G BASE-DR4" that realizes up to 400Gbps with "PSM4" that bundles 4 pairs of 53.125G "PAM-4"Copyright © 2018 Impress Corporation. All rights reserved.