SUPPLEMSUPPLEMENT


THE ININTERNET of TH NG SHINGI


Vestberg (Ericsson Jon Iwata (IBM) , an DaveEva s (Csc wo)were alll bullish in the early 2010s,10s, typically estimating that more than 50 billion connected devices would be in use by 2020.. Since thenth these estimates have become more realistic. But today even the most conservative IoT predictions claim that there will be more than 20 billion connected devices by 2020.. Here Benjjamiin Jordan,AltiumLLC Altium LLC oks at how‘things’ are evol


typcallye tim tin th theseest th


PUTT ty


th re ks


t’s hard to design boards right the first time in every respect. But in spite of the increasing difficulty with cluttered channel space in the wireless spectrum, we are pretty good at


I Figure 1:


designing for EMC and signal integrity as well. Yet board designers still have to fight to reduce design time and respins due to other problems. The real bottlenecks are exacerbated by the nature of IoT devices, with increasing complexity and density.


What can we do to increase design throughput? What about Automation? PCB designers and engineers will always say the same thing: “An autorouter makes a mess and cannot do what I can do. I will never use an autorouter!” For all but the most simple designs that have no impedance control, no sensitive analogue circuits, and no high-speed topology requirements; our experience of routing aut omation in times past has not been great.


DEEP EARNING FOR ED DEEP LEARNING FOR EDA


AI and Deep Learning are the crux of this matter. With all those complex boards to design, we need automation to help with component placement and routing of the PCB. We need it whether we want it or not. But it has to work. I say AI and Deep Learning because we are at a point in time with technology that can rapidly accelerate the learning of routing and placement algorithms. It’s been tried several times before. Neural networks for example have been used extensively in the development of 1990s era autorouting tools, such as Neuroroute. Neural networks are


connectivity or relationship graphs that are “trained” by feeding themexample stimuli and outputs. The idea is that the neural net “learns” the human way of routing PCBs and crea tes a set of parameters that cause it to mimic how a human would do the ro Neuroroute (Figure 1) w


as arguably uting.


the first even neural-net based topological autorouter developed. Acquired by Protel (now Altium) in the late 1990s, it formed the base engine


S6 S6 DECEMBER JANUAR 201 ECEMBER/JANUARY 2018 | ELEC ELECTRONICS


of what became the Situs autorouter that is in all the company’s PCB tools. It had great promise, and for 90s-era PCB design specs it does well. The problemis that it was not able to adapt to the much finer pitches and BGAs that hit PCB designers in the early 2000s, because the base set of PCBs used to “train” it was very limited.


Since then a number of largely improved routing tools have been


developed. The topological router can be Figure 3:Figure 3:


Situs results with 2-layer Situs results with 2-layer SMT board (Source: Altium L


MT board (Source: Altium LLC 2016)


2016)


Neuroroute user manual front page (Source: Altium LLC 2017)


al looksathow b


PUTTINGTHE ‘ HansVestberg


reth


stim tesh vebecom llbe


ve ec reth


TTING THE ‘A’BAC stberg (Ericsson),, JonIwa


rerea st llio co


re volvingacro


ealistic.B t tod ect


cte


wata(IB ) ,andDaveEv llio co


cte


BACK I TO EDA FOR I lb llishinth


ACKIN TOEDA vans (Ciisco) were


ow hings’areevreevoving across this digital landscape ross thi digital ndscape


odayeve evcesb ca


SPECCTRA by mappi based” routers (Figu compared with tradi


tional “shape


ng the board space re 2) such as


and obstacles using polygons, and applying different path costs (for example, adding vias versus going the long way around the board), to


optimise the routing while obeying the design rules. For simple boards like the one shown in Figure 3, it can be set u p to do a decent job.


The current generation of routing


automata experts recognise the need for more user-interactive automation. The ActiveRoute tool in AltiumDesigner uses path-cost optimised topological routing driven by user selection and guidance. This generation of routing automation is a step forw


rward but it won’t be enough to Figure 2: Figure 2: Rectilinear vs. Rectilinear vs vs.


ttopological routing (Source: Altium L 2017)


opological routing (Source: Altium LLC 2017)


carry us into the future of IoT.


Neural Net based topological routing engines need to be more powerful. The only way we can improve themto th e point of true human-like results for every desktop is through Deep Learning. Deep Learning is a fancy term for


expanding the data set that teaches the artificial intelligence engine called a Deep Neural Network, applying that expansive set of stimuli and responses to develop the machine’s behaviour. What if the neural net routing


automation had a practically unlimited set of PCB designs to train it? What if , each time you design a new PCB in your EDA tool, the EDA tool was able to learn your moves and begin to help you by anticipating how routed, or even how


moving some parts the board could be


on the board (within your constraints of course) could drastically reduce routing time and layer count? What if you could leverage and


contribute to (voluntarily) a hive-mind of engineers and PCB designers anonymously to train the world' s


greatest routing engine? These are real questions which will be answered over the coming months and years.


Altium


www.altium.com T: 01920 876250


www.altium.com / ELECTRONICS ELECTRONICS


cted evceswo ev


w


venthem st by


bein rv n by


st conserva ive T H re


vativ S


veIoTpred t nJord


edictions clai LC


DAFORIO T rly


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56