WEBVTT 1 00:00:00.080 --> 00:00:04.080 Welcome to this deep dive. Today. We're digging into analog 2 00:00:04.200 --> 00:00:07.080 circuit design, specifically around the year twenty ten. 3 00:00:07.280 --> 00:00:09.679 Yeah, we've got this great collection of discussions from the 4 00:00:09.720 --> 00:00:11.199 ACD workshop back. 5 00:00:11.039 --> 00:00:13.439 Then, and our mission really is to pull out the 6 00:00:13.519 --> 00:00:17.879 key developments the future ideas across well three main areas. 7 00:00:18.199 --> 00:00:22.559 That's robust design, Sigma delta converters, and RFID exactly. 8 00:00:22.640 --> 00:00:26.120 Think of this as like your shortcut to understanding some 9 00:00:26.120 --> 00:00:29.320 pretty complex stuff without getting totally bogged down. You might 10 00:00:29.359 --> 00:00:30.879 find some surprising things here. 11 00:00:30.960 --> 00:00:35.439 It's a fascinating snapshot. Really, twenty ten was a pivotal moment. 12 00:00:35.759 --> 00:00:38.119 Nanoscale CMOS was becoming real. 13 00:00:38.000 --> 00:00:41.359 Which open doors but also threw up new roadblocks totally. 14 00:00:41.520 --> 00:00:44.320 So we'll look at how engineers were tackling robustness in 15 00:00:44.359 --> 00:00:47.439 that new world, the cool stuff happening in data conversion 16 00:00:47.479 --> 00:00:50.759 with signa deltas, and you know, the whole RFID universe 17 00:00:50.759 --> 00:00:51.479 started to expand. 18 00:00:51.520 --> 00:00:54.560 Perfect. Okay, let's start with robust design. That was clearly 19 00:00:54.600 --> 00:00:58.079 a huge theme. Everyone was pushing for smaller features, you know, 20 00:00:58.159 --> 00:01:00.200 ninety nimeters even smaller, and. 21 00:01:00.159 --> 00:01:03.399 That unlocked amazing things. But yeah, the obstacles were real. 22 00:01:03.759 --> 00:01:06.480 It's where the physics, the tiny atomic level stuff really 23 00:01:06.519 --> 00:01:08.120 started hitting circuit design hard. 24 00:01:08.239 --> 00:01:11.200 So engineers couldn't just use the old bulk models anymore. 25 00:01:11.200 --> 00:01:15.359 Pretty much, the designs got way more complex, which stretched 26 00:01:15.359 --> 00:01:18.000 out time to market. And then there's the variability. The 27 00:01:18.040 --> 00:01:20.840 transitionors themselves just weren't perfectly identical. 28 00:01:20.439 --> 00:01:23.599 Anymore, leading to mismatch, lower yields. 29 00:01:23.359 --> 00:01:27.959 Exactly, and reliability over time became a bigger worry too, 30 00:01:28.120 --> 00:01:34.560 things like hot carriers NbTi. Basically the circuits degrading. 31 00:01:34.159 --> 00:01:37.760 Negative bias, temperature instability. That one sounds like a nightmare 32 00:01:37.799 --> 00:01:41.040 for designers. And you mentioned even the physical layout got trick. 33 00:01:40.840 --> 00:01:46.319 Here, Oh yeah, more rules, tighter constraints, and these optical 34 00:01:46.359 --> 00:01:49.159 proximity effects started messing with the shapes they were trying 35 00:01:49.200 --> 00:01:50.040 to print on the silicon. 36 00:01:50.280 --> 00:01:55.040 So the manufacturing process itself added challenges. But robustness wasn't 37 00:01:55.079 --> 00:01:57.359 just about nanoscale, was it. There was this push for 38 00:01:57.480 --> 00:01:59.640 circuits and really tough environments too. 39 00:02:00.040 --> 00:02:03.560 Absolutely a big one was high temp, high voltage electronics, 40 00:02:03.799 --> 00:02:06.799 I think electric and hybrid vehicles which were just starting 41 00:02:06.799 --> 00:02:10.280 to take off. Their challenge there was making sure everything 42 00:02:10.280 --> 00:02:14.240 worked together reliably. The silicon, the packaging the materials. It 43 00:02:14.319 --> 00:02:19.919 needed complex simulations, thermal electric stuff, reliability calculations, especially for 44 00:02:19.960 --> 00:02:21.360 the smart power modules. 45 00:02:21.080 --> 00:02:24.000 And they were making real progress. I saw mentions of 46 00:02:24.120 --> 00:02:26.840 igbt's handling pretty extreme temperatures. 47 00:02:26.919 --> 00:02:28.960 Yeah, junction temps up to one hundred and seventy five 48 00:02:29.000 --> 00:02:32.039 degree C, sometimes even near two hundred C in short 49 00:02:32.039 --> 00:02:32.719 circuit tests. 50 00:02:32.800 --> 00:02:37.000 Wow, that's incredibly hot. Must have involved some serious trade 51 00:02:37.000 --> 00:02:39.360 offs between cooling and performance for sure. 52 00:02:39.599 --> 00:02:42.680 And this need for high temp electronics wasn't just in 53 00:02:42.719 --> 00:02:46.120 the main powertrain. It was sensors, motors all over the. 54 00:02:46.159 --> 00:02:48.800 Vehicle, different power levels, different integration approaches. 55 00:02:48.879 --> 00:02:51.639 Right, monolithic versus multi chip modules. Yeah, that was a 56 00:02:51.639 --> 00:02:54.759 big decision point for reliability too. The whole thing was 57 00:02:54.800 --> 00:02:58.599 guided by tough auto industry standards like AECQ one oh one. 58 00:02:58.800 --> 00:03:01.400 It makes sense make sure your car works reliably hot 59 00:03:01.439 --> 00:03:05.759 or cold, Okay. Another hostile environment mentioned was radiation space 60 00:03:05.800 --> 00:03:08.439 for example. That sounds totally different it is. 61 00:03:08.759 --> 00:03:12.240 You've got cumulative damage over time and also these single 62 00:03:12.280 --> 00:03:13.639 hits from hy energy particles. 63 00:03:13.719 --> 00:03:14.520 Nasty stuff. 64 00:03:14.759 --> 00:03:17.280 But what's interesting is that researchers were finding ways to 65 00:03:17.400 --> 00:03:22.400 use standard commercial CMOS. Turns out with the gate oxides 66 00:03:22.439 --> 00:03:26.879 getting thinner, like around five nanometers. Yeah, the threshold voltage 67 00:03:26.879 --> 00:03:30.719 shift from radiation became almost negligible for a lot of uses. 68 00:03:30.960 --> 00:03:35.840 Really, so you didn't always need special expensive rad hard processes. 69 00:03:36.159 --> 00:03:39.080 Not always. It wasn't perfect, mind you. There were still 70 00:03:39.120 --> 00:03:41.840 issues like with the STI oxide leakage varying a lot. 71 00:03:41.919 --> 00:03:43.479 Okay, so still burdles. 72 00:03:43.439 --> 00:03:47.840 But they developed clever layout tricks, things like ring sources 73 00:03:48.080 --> 00:03:52.319 guard rings basically protective structures around their transistors. 74 00:03:51.759 --> 00:03:53.159 To shield them a bit sort of. 75 00:03:53.280 --> 00:03:56.439 Yeah, improve the tolerance to the total radiation dose. And 76 00:03:56.439 --> 00:03:59.120 then for those single particle hits, especially in memory. 77 00:03:58.879 --> 00:04:01.439 Like bitflips and surround exactly. 78 00:04:01.120 --> 00:04:04.199 They use techniques like redundancy special cell designs like the 79 00:04:04.199 --> 00:04:08.120 Whittaker cell DICE architectures, although DICE E was showing limits 80 00:04:08.159 --> 00:04:10.000 in newer technologies. 81 00:04:09.319 --> 00:04:13.319 And things like triple modular redundancy error correction codes. 82 00:04:13.400 --> 00:04:16.800 Yeah, TMR and EEDAC were key strategies too, especially for 83 00:04:16.879 --> 00:04:18.839 protecting against those single event upsets. 84 00:04:19.079 --> 00:04:23.360 Okay, some smart design to fight back against cosmic rays fascinating. 85 00:04:23.920 --> 00:04:29.639 And the third tough environment electromagnetic compatibility EMC. 86 00:04:29.519 --> 00:04:31.199 Especially for power switches. 87 00:04:31.040 --> 00:04:34.279 Absolutely critical. You pack more and more electronics into a car. 88 00:04:34.319 --> 00:04:36.959 They can't interfere with each other. That's a safety thing, right. 89 00:04:37.279 --> 00:04:40.439 So the trend was towards simulating this stuff earlier, using 90 00:04:40.480 --> 00:04:43.240 techniques like direct power injection testing, but doing it in 91 00:04:43.240 --> 00:04:44.519 the simulation world to. 92 00:04:44.519 --> 00:04:46.160 Predict problems before building. 93 00:04:45.879 --> 00:04:49.319 The chip exactly. And a key technique was optimizing the 94 00:04:49.360 --> 00:04:52.759 impedance of the pins on those smart power switches to 95 00:04:52.800 --> 00:04:55.759 make them less susceptible. So designing the chip to be 96 00:04:55.800 --> 00:05:00.319 a good neighbor in that noisy electrical environment makes sense. Okay, 97 00:05:00.439 --> 00:05:04.000 shifting focus a bit, let's talk reliability more broadly in 98 00:05:04.160 --> 00:05:08.879 nanoscale cmos, not just extreme conditions but everyday stress. 99 00:05:09.160 --> 00:05:13.000 Right, So a big part was modeling that long term 100 00:05:13.000 --> 00:05:15.759 degradation hot carriers and BTI. 101 00:05:15.879 --> 00:05:17.319 Again, how do they model it. 102 00:05:17.399 --> 00:05:21.319 They'd simulate how transistors behaved under typical signal stress and 103 00:05:21.360 --> 00:05:25.120 then use models to kind of extrapolate how much they'd 104 00:05:25.120 --> 00:05:26.439 degrade over the chip's. 105 00:05:26.120 --> 00:05:29.560 Lifetime, like accelerated aging, but in simulation pretty much. 106 00:05:29.600 --> 00:05:31.600 Yeah, helps find the weak spots early. 107 00:05:31.759 --> 00:05:35.240 Smart And what about dealing with manufacturing variations, because no 108 00:05:35.319 --> 00:05:37.120 two chips are ever exactly the same? 109 00:05:37.240 --> 00:05:40.319 True? That process variability was a big headache. One approach 110 00:05:40.439 --> 00:05:41.959 was post fabrication. 111 00:05:41.519 --> 00:05:43.600 Calibration, tuning the chip after it's made. 112 00:05:43.759 --> 00:05:47.879 Yeah, Like the example was using switching sequence post adjustment 113 00:05:47.920 --> 00:05:49.800 on a DC. Yeah, they could tune it up and 114 00:05:49.800 --> 00:05:52.519 actually use less chip area to get the same accuracy. 115 00:05:52.600 --> 00:05:56.000 Clever, And then there were techniques for adapting while the circuit. 116 00:05:55.680 --> 00:05:59.839 Is running right runtime self adaptation. The example was a 117 00:05:59.839 --> 00:06:04.079 lot N driver with extra backup transistors like spares kind of. 118 00:06:04.519 --> 00:06:07.160 It had a monitor circuit that could switch in replacements 119 00:06:07.439 --> 00:06:10.360 if others started to degrade, keeps performance. 120 00:06:09.920 --> 00:06:11.839 Stable, but that takes up more space. 121 00:06:11.959 --> 00:06:15.439 I guess, yeah, there's always a trade off area versus resilience. 122 00:06:15.560 --> 00:06:19.439 Okay, Now this was also when hike metal gate technology 123 00:06:19.519 --> 00:06:22.360 was coming in. How did that change the variability picture? 124 00:06:22.839 --> 00:06:26.639 It was a mixed bag. It helps with some variability 125 00:06:26.639 --> 00:06:30.639 sources like random dope fluctuations, but it introduced new ones 126 00:06:30.839 --> 00:06:34.959 like the high material wasn't perfectly uniform, granularity issues and 127 00:06:35.079 --> 00:06:36.920 variations in the metal gates work function. 128 00:06:37.160 --> 00:06:39.839 So solved one problem created another. 129 00:06:39.639 --> 00:06:42.480 A bit yeah, and things like random dope ins line 130 00:06:42.600 --> 00:06:47.959 is roughness, polygrain boundaries. They still caused threshold voltage variations, 131 00:06:48.279 --> 00:06:50.439 especially for the really short channel transistors. 132 00:06:50.519 --> 00:06:52.160 The smallest ones were most sensitive. 133 00:06:52.279 --> 00:06:54.920 They really were needed very careful design, and all this 134 00:06:55.000 --> 00:06:58.480 complexity really flowed down into the physical design, the actual layout. 135 00:06:58.600 --> 00:07:01.920 I bet more rules, tighter tolerances must have been tough. 136 00:07:02.040 --> 00:07:06.240 It was. Designers maybe had less direct control over layout, 137 00:07:06.439 --> 00:07:11.639 and those optical proximity effects corner rounding slight misalignments, they 138 00:07:11.680 --> 00:07:13.199 caused matching errors. 139 00:07:13.040 --> 00:07:16.079 So tiny manufacturing imperfections mattered more. 140 00:07:16.519 --> 00:07:20.560 A lot. More strategies involve things like using only certain 141 00:07:20.600 --> 00:07:24.680 device widths for regularity. Considering stress from metal wires. 142 00:07:24.319 --> 00:07:27.240 Above, it sounds like circuit design and physical layout had 143 00:07:27.279 --> 00:07:29.240 to become much more tightly integrated. 144 00:07:29.319 --> 00:07:33.160 Absolutely planning the physical aspects early was key to avoiding 145 00:07:33.199 --> 00:07:33.920 problems later. 146 00:07:34.040 --> 00:07:38.639 Okay, last point on robust design, packaging and interconnects, especially 147 00:07:38.639 --> 00:07:41.639 for those high power automotive modules where silicon meets the 148 00:07:41.680 --> 00:07:42.920 real world exactly. 149 00:07:43.079 --> 00:07:46.920 Multi chip modules were common in EVSAGVS, and a huge 150 00:07:46.920 --> 00:07:51.199 issue was mechanical stress. Stress from what different materials expanding 151 00:07:51.199 --> 00:07:53.800 and contracting at different rates when the temperature changes, you 152 00:07:53.839 --> 00:07:57.439 know a CTE mismatch. Silicon expands differently than the ceramic 153 00:07:57.480 --> 00:07:59.040 substrate or the base plate. 154 00:07:59.000 --> 00:08:03.000 Ah sot up and cooling down put strain on everything. 155 00:08:02.600 --> 00:08:05.360 Big time, especially over the wide temperature range. Cars operate 156 00:08:05.439 --> 00:08:08.360 in like netteck of forty c up to really high temps. 157 00:08:08.160 --> 00:08:11.279 And that affects the connections soldering wire bonds. 158 00:08:11.519 --> 00:08:14.560 Definitely higher temps were tough on wire bonds they could 159 00:08:14.560 --> 00:08:17.199 lift off. They did lots of power cycling tests to 160 00:08:17.279 --> 00:08:21.279 check durability. Even the direct bond copper structures felt the 161 00:08:21.319 --> 00:08:22.920 strain from thermal cycles. 162 00:08:23.199 --> 00:08:25.600 Sounds like material science was just as crucial as the 163 00:08:25.680 --> 00:08:26.240 chip design. 164 00:08:26.240 --> 00:08:29.279 Here. It really was, and they were already looking ahead 165 00:08:29.319 --> 00:08:33.399 exploring things like metal carbon nanotube composites from what better 166 00:08:33.480 --> 00:08:37.480 thermal interface materials maybe even metallization. C and ts have 167 00:08:37.639 --> 00:08:41.960 amazing thermal and electrical conductivity way better than copper. Could 168 00:08:42.000 --> 00:08:44.480 lead to even more robust, high temp power electronics down 169 00:08:44.519 --> 00:08:44.840 the road. 170 00:08:45.240 --> 00:08:48.480 Interesting glimpse into the future there. Okay, that wraps up 171 00:08:48.559 --> 00:08:52.000 robust design. Let's switch gears to our second topic. Sigma 172 00:08:52.000 --> 00:08:56.480 delta converters essential for higher precision analog to digital conversion. 173 00:08:56.799 --> 00:08:59.000 Yeah, and the focus here was oft and on getting 174 00:08:59.080 --> 00:09:03.879 higher resolution your bandwidth but using less power. Constant innovation. 175 00:09:04.080 --> 00:09:07.679 One area you highlighted was noise coupled delta sigma ADCs 176 00:09:08.399 --> 00:09:09.399 using noise to help. 177 00:09:09.720 --> 00:09:13.000 That sounds backward, it does, right, but the idea is 178 00:09:13.039 --> 00:09:16.360 really clever. You take the quantization noise generated in one 179 00:09:16.399 --> 00:09:17.799 part of the modulator. 180 00:09:17.320 --> 00:09:18.679 The unwanded stuff. 181 00:09:18.440 --> 00:09:21.080 Right, and you strategically inject it somewhere else in loop. 182 00:09:21.639 --> 00:09:25.320 In certain architectures like feed forward ones, this could actually 183 00:09:25.360 --> 00:09:28.120 boost performance without needing extra op amps. 184 00:09:28.279 --> 00:09:29.559 Huh, what's the catch? 185 00:09:30.200 --> 00:09:32.879 Well, you usually sacrifice a bit of the maximum input 186 00:09:33.000 --> 00:09:36.399 signal range you can handle without distortion, so choosing the 187 00:09:36.519 --> 00:09:40.279 quantizer resolution carefully is key. Usually need like one extra 188 00:09:40.360 --> 00:09:41.919 bit than the modulator. 189 00:09:41.559 --> 00:09:44.759 Order, so you're kind of using the noise against itself, 190 00:09:44.799 --> 00:09:47.480 but you can't push it too hard. Were there different 191 00:09:47.480 --> 00:09:49.320 ways to do this noise coupling. 192 00:09:49.080 --> 00:09:54.320 Oh yeah, split modulators, noise canceling structures, noise coupled, time interleaved. 193 00:09:55.039 --> 00:09:58.360 That last one. NCTI looked pretty promising. Why is that 194 00:09:58.399 --> 00:10:01.399 it had this enhancement factor that improve things, and time 195 00:10:01.440 --> 00:10:04.639 interleaving made it simpler to implement. Plus it was less 196 00:10:04.639 --> 00:10:08.200 sensitive to mismatch errors between channels because it's suppressed noise 197 00:10:08.240 --> 00:10:09.559 around the niquist frequency. 198 00:10:09.799 --> 00:10:11.840 But there was still a trade off sq and R 199 00:10:11.960 --> 00:10:13.000 versus robustness. 200 00:10:13.240 --> 00:10:16.519 Always trade offs and analog design. But yeah, different ways 201 00:10:16.519 --> 00:10:17.639 to leverage the concept. 202 00:10:17.879 --> 00:10:22.840 Okay. Another interesting idea, very low over sampling ratio sigma 203 00:10:22.879 --> 00:10:26.120 deltas isn't oversampling like the whole point of sigma delta. 204 00:10:26.279 --> 00:10:29.240 It usually is, but at very low osrs they start 205 00:10:29.240 --> 00:10:33.519 competing head to head with niquist ADCs like pipeline ADCs. 206 00:10:33.559 --> 00:10:36.799 Really yeah, the sources suggest there's a crossover point where 207 00:10:37.120 --> 00:10:40.200 higher order sigma deltas can actually be lower order ones. 208 00:10:40.519 --> 00:10:43.679 Even with low osrs. You still get noise reduction even 209 00:10:43.720 --> 00:10:45.000 with an OSR of to say. 210 00:10:45.080 --> 00:10:48.000 Three, surprising, so you're relying more on the heavy duty 211 00:10:48.039 --> 00:10:51.080 noise shaping even with fewer samples. How did they stack 212 00:10:51.120 --> 00:10:53.279 up against pipelines on power? That's often critical? 213 00:10:53.399 --> 00:10:57.200 Good question. Pipelines sometimes use these half delaying stages, which 214 00:10:57.240 --> 00:11:00.399 could be less power efficient than masah shigma delta using 215 00:11:00.399 --> 00:11:04.200 full period integrators. Okay, but people were working on making 216 00:11:04.200 --> 00:11:08.519 pipelines more efficient too, swushowpams, sharing amplifiers, things like that, 217 00:11:08.840 --> 00:11:12.480 and incremental ADCs also perform will low osrs getting good 218 00:11:12.519 --> 00:11:14.279 resolution because they can handle bigger signals. 219 00:11:14.480 --> 00:11:17.600 It feels like the lines between architectures were blurring a 220 00:11:17.600 --> 00:11:21.000 bit at these low osrs. Everyone borrowing ideas. 221 00:11:21.240 --> 00:11:23.759 Kind of yeah, optimizing for performance and power. 222 00:11:24.720 --> 00:11:30.080 Now what about compartor based switched capacitor sigma deltas using 223 00:11:30.159 --> 00:11:32.840 comparitors instead of op ams? Why do that. 224 00:11:32.960 --> 00:11:37.720 Big advantage in nanoscale cmos Designing good op amps with 225 00:11:37.799 --> 00:11:42.279 low supply voltage just gets really hard. Comparators can be simpler, faster, 226 00:11:42.440 --> 00:11:43.559 potentially lower power. 227 00:11:43.679 --> 00:11:44.159 Makes sense. 228 00:11:44.320 --> 00:11:47.759 They explore different ways to build integrators with comparators. Pseudo 229 00:11:47.799 --> 00:11:51.480 differential looked good for high speed. It cleverly turns comparator 230 00:11:51.519 --> 00:11:54.240 delay into a common mode error, which is easier to 231 00:11:54.279 --> 00:11:54.919 deal with later. 232 00:11:55.159 --> 00:11:57.679 So you trade the op AMS precision for the comparator 233 00:11:57.759 --> 00:12:00.720 speed and power benefits, then clean up them us elsewhere. 234 00:12:00.799 --> 00:12:03.360 What about that comparator delay? Did it cause problems? 235 00:12:03.600 --> 00:12:07.440 It could cause some output voltage overshoot, maybe a DC offset, 236 00:12:07.759 --> 00:12:10.080 But since comparators are open loop you don't have the 237 00:12:10.080 --> 00:12:13.720 same stability worries as with opams and feedback. The key 238 00:12:13.759 --> 00:12:17.279 thing is just the delay time itself, which interestingly didn't 239 00:12:17.279 --> 00:12:19.279 depend much on the filter capacitor sizes. 240 00:12:19.360 --> 00:12:21.320 Okay, and noise was still a factor obviously. 241 00:12:21.480 --> 00:12:25.679 Oh yeah, thermal noise from switches, reference voltages. You still 242 00:12:25.679 --> 00:12:27.679 had to analyze all that to figure out the SNR, 243 00:12:28.519 --> 00:12:31.120 but it was definitely a viable path, especially for low power. 244 00:12:31.200 --> 00:12:34.559 Then there was the really different idea VCO based ADCs 245 00:12:35.399 --> 00:12:36.960 voltage controlled oscillators. 246 00:12:37.080 --> 00:12:39.720 Yeah, pretty novel. The VCO acts like a voltage to 247 00:12:39.799 --> 00:12:42.559 time converter. Combine that with the time to digital converter 248 00:12:43.039 --> 00:12:45.639 and you get noise shaping, even mismatch shaping. 249 00:12:46.159 --> 00:12:48.399 But VCOs aren't perfectly linear, are they. 250 00:12:48.559 --> 00:12:51.519 That's the main challenge. The relationship between input voltage and 251 00:12:51.519 --> 00:12:54.279 output frequency isn't straight. The trick was to use the 252 00:12:54.320 --> 00:12:56.879 phase of the VCO output as the key variable. 253 00:12:56.879 --> 00:12:57.200 Phase. 254 00:12:57.240 --> 00:13:00.600 Okay, theoretically that could get you really high SNDR over 255 00:13:00.639 --> 00:13:04.440 wide bandwidths, and the prototypes mentioned showed promising results backing 256 00:13:04.480 --> 00:13:05.000 up the theory. 257 00:13:05.279 --> 00:13:09.240 So encoding voltage into frequency or phase a totally different 258 00:13:09.240 --> 00:13:13.919 way to digitize cool. Lastly, wide band continuous time multi 259 00:13:13.919 --> 00:13:17.039 bit delta sigmas using more than one bit in the quantizer. 260 00:13:17.919 --> 00:13:22.159 That has advantages, big advantages. Multibit contizers ease the requirements 261 00:13:22.159 --> 00:13:24.720 on the op amps, especially their SLEW rate and they 262 00:13:24.759 --> 00:13:26.879 reduce the out of band quantization. 263 00:13:26.480 --> 00:13:28.120 Noise leading to lower power. 264 00:13:28.360 --> 00:13:32.000 Often yes, but the trade off is the DK in 265 00:13:32.039 --> 00:13:35.320 the feedback loop becomes more complex. You need dynamic element 266 00:13:35.360 --> 00:13:37.000 matching DEM to keep it. 267 00:13:36.879 --> 00:13:38.919 Linear, right, got to make sure that this is accurate. 268 00:13:39.120 --> 00:13:43.519 But despite that complexity, low OSR combined with multibit quantizers 269 00:13:43.519 --> 00:13:47.679 and DEM algorithms like DWA was becoming really popular for 270 00:13:47.759 --> 00:13:48.600 wideband stuff. 271 00:13:48.960 --> 00:13:52.639 So trading some over sampling for multibit benefits and using 272 00:13:52.679 --> 00:13:56.159 DEM to fix the DT sounds like a smart compromise exactly. 273 00:13:56.679 --> 00:14:00.559 And the performance figures reported were impressive high dynamic range 274 00:14:00.960 --> 00:14:05.240 high SNDR. They were using cascaded architectures, high speed decks, 275 00:14:05.720 --> 00:14:07.000 really pushing the envelope. 276 00:14:07.039 --> 00:14:10.279 Clearly, Sigma delta was a hot area for innovation back then. 277 00:14:10.360 --> 00:14:13.720 Absolutely just relentless efforts to get better performance, less power, 278 00:14:13.759 --> 00:14:15.679 more bandwidth, lots of creativity. 279 00:14:15.720 --> 00:14:20.840 Okay, final topic, r FID radio frequency identification. Around twenty ten, 280 00:14:20.879 --> 00:14:23.120 this was really starting to break out of niche uses, 281 00:14:23.159 --> 00:14:23.559 wasn't it. 282 00:14:23.559 --> 00:14:26.120 It really was. You know, it started way back tracking 283 00:14:26.320 --> 00:14:29.639 nuclear stuff than cows, but by twenty ten people were 284 00:14:29.639 --> 00:14:31.840 seriously looking at it for much broader things like what 285 00:14:31.960 --> 00:14:35.240 tagging High value items was a big one mentioned, fashion, 286 00:14:35.480 --> 00:14:40.799 fancy goods, electronics, huge potential for better inventory management. The 287 00:14:40.919 --> 00:14:43.840 key was the move towards open standards that was crucial for. 288 00:14:43.720 --> 00:14:47.639 Getting scale from cows to couture. Sounds like the market 289 00:14:47.720 --> 00:14:51.480 was poised for growth, but maybe some hurdles remained. 290 00:14:51.559 --> 00:14:55.559 For sure, technical issues like getting reliable reads in different environments, 291 00:14:55.720 --> 00:14:58.919 the cost both tags and readers, and figuring out who 292 00:14:58.960 --> 00:15:00.720 pays and who benefits the supply chain. 293 00:15:00.840 --> 00:15:02.440 Yeah, the business side, but the. 294 00:15:02.320 --> 00:15:06.120 Long term vision was much lower costs readers maybe two hundred, 295 00:15:06.360 --> 00:15:08.919 passive tags just a few cents. They even talk about 296 00:15:08.960 --> 00:15:12.759 chipless tags under a cent that was further out but revolutionary. 297 00:15:12.799 --> 00:15:15.440 If it happened, Wow, that would make it truly everywhere. 298 00:15:15.480 --> 00:15:18.039 Retail and consumer goods were the big targets. 299 00:15:17.759 --> 00:15:20.559 Expected to be the largest market. Yeah. UHF was seen 300 00:15:20.600 --> 00:15:23.799 as the main frequency for tracking objects, though HF had 301 00:15:23.840 --> 00:15:26.000 its niches like libraries. 302 00:15:25.600 --> 00:15:28.679 And the trend was better performance, lower cost, making it 303 00:15:28.799 --> 00:15:29.799 essential for business. 304 00:15:30.039 --> 00:15:32.840 That was the expectation, especially in Europe. But for that 305 00:15:32.919 --> 00:15:36.039 kind of widespread use, standards are absolutely vital. 306 00:15:35.960 --> 00:15:41.360 Right, and the standardization effort had started way back isoie. 307 00:15:40.960 --> 00:15:44.039 Yeah, JTC one, SC thirty one, double G four formed 308 00:15:44.039 --> 00:15:47.200 A ninety seven. Their goal was a common protocol for 309 00:15:47.279 --> 00:15:50.639 interoperability across the supply chain or really broad scope. 310 00:15:50.639 --> 00:15:55.279 But before that it was more fragmented application specific standards mostly. 311 00:15:55.000 --> 00:16:00.480 Yeah ANILID road tolls. Not much focus on broad tech 312 00:16:00.480 --> 00:16:04.279 based standards covering different frequencies, so maybe ANSI. In the US. 313 00:16:04.639 --> 00:16:07.799 A lot of early tech was proprietary, leading to closed systems, 314 00:16:08.120 --> 00:16:11.000 and UHF had some regulatory issues too, potential overlap with 315 00:16:11.000 --> 00:16:11.799 mobile phones in. 316 00:16:11.720 --> 00:16:13.960 Some places, so a bit messy. Initially, it sounds like 317 00:16:14.080 --> 00:16:16.720 RFID standards had some catching up to do compared to 318 00:16:16.759 --> 00:16:17.679 say barcodes. 319 00:16:17.759 --> 00:16:20.679 That's a key point from the sources. The barcode community 320 00:16:20.720 --> 00:16:23.360 had a deep understanding of applications in tech when their 321 00:16:23.399 --> 00:16:27.240 standards evolved. RFID was maybe ten twenty years behind in 322 00:16:27.240 --> 00:16:30.000 that sense, starting almost from scratch in terms of consensus. 323 00:16:29.679 --> 00:16:31.519 Must have been hard getting everyone on the same page 324 00:16:31.519 --> 00:16:32.799 with all the different technologies. 325 00:16:32.919 --> 00:16:35.360 Definitely, the first phase of ISO eighteen thousand tried to 326 00:16:35.360 --> 00:16:38.159 set up rules, maybe one protocol pre frequency, but there 327 00:16:38.159 --> 00:16:40.440 were just so many competing proprietary candidates. 328 00:16:40.639 --> 00:16:44.039 What about the application side EPC Global had groups working 329 00:16:44.039 --> 00:16:44.279 on that. 330 00:16:44.399 --> 00:16:47.759 Yeah, those were crucial. They worked on things like lookup services, yeah, 331 00:16:47.799 --> 00:16:52.399 the discovery service, supply chain tracking models, and importantly security 332 00:16:52.440 --> 00:16:55.840 and privacy analysis, building prototypes. 333 00:16:55.159 --> 00:16:58.799 So practical implementation, not just the tech specs right. 334 00:16:58.919 --> 00:17:03.480 And business groups focused on sectors like anti counterfeiting, pharma, apparel, food, 335 00:17:04.039 --> 00:17:06.640 running pilot projects to show the value any. 336 00:17:06.440 --> 00:17:09.039 Cool examples of early smart object systems. 337 00:17:09.119 --> 00:17:12.400 I mentioned smart shells for bookstores, remote servicing for heavy 338 00:17:12.400 --> 00:17:17.160 equipment using RFID to spot issues or manage parts, showing 339 00:17:17.160 --> 00:17:19.319 that potential beyond just tracking, and. 340 00:17:19.200 --> 00:17:22.480 The performance requirements for logistics were getting pretty demanding. Read 341 00:17:22.559 --> 00:17:24.000 range speed. 342 00:17:23.680 --> 00:17:26.359 Yeah, reading up to nine meters, reading whole pallettes fast 343 00:17:26.400 --> 00:17:28.440 like half a second to two seconds, needing new one 344 00:17:28.519 --> 00:17:32.240 hundred percent accuracy which meant more power, higher data rates, 345 00:17:32.359 --> 00:17:34.839 and people were already thinking about adding sensors. 346 00:17:34.519 --> 00:17:37.799 To tag sensors like temperature shock. 347 00:17:37.720 --> 00:17:42.319 Exactly, temperature, humidity pressure. Moving beyond just ID to collecting 348 00:17:42.400 --> 00:17:45.160 environmental data that needed more power and data capacity, of. 349 00:17:45.200 --> 00:17:48.119 Course opens up a lot more possibilities. Okay, let's seme 350 00:17:48.119 --> 00:17:52.480 in on the super small stuff ultra small RFID chips. 351 00:17:52.559 --> 00:17:56.839 The hon chips, tiny chips and everyday objects. Sounds futuristic 352 00:17:56.960 --> 00:17:57.680 even now. 353 00:17:57.599 --> 00:18:00.519 It really does. The goal was linking physical things to 354 00:18:00.640 --> 00:18:05.400 network information. Small cheap chips for anti counterfeiting product tracking. 355 00:18:05.200 --> 00:18:08.680 Like that chip at the IGXPO point four milimeter square Yeah, two. 356 00:18:08.519 --> 00:18:10.440 Point four or five getter herds one hundred and twenty 357 00:18:10.440 --> 00:18:13.119 eight bit ID and the trend was even smaller like 358 00:18:13.160 --> 00:18:15.920 Hatachi's sub point one millimeter chips. 359 00:18:16.079 --> 00:18:18.599 Incredible like electronic dust. What's inside them. 360 00:18:18.480 --> 00:18:20.839 Basically ROM for the ID one twenty eight bits like 361 00:18:20.880 --> 00:18:23.640 an IPv six address conceptually, and a rectifier circuit to 362 00:18:23.640 --> 00:18:26.839 harvest RF energy for power. Early ones skipped anti collision 363 00:18:26.839 --> 00:18:27.720 to save space. 364 00:18:27.519 --> 00:18:28.079 And the antenna. 365 00:18:28.240 --> 00:18:29.920 Some had it embedded right on the chip for two 366 00:18:29.920 --> 00:18:32.440 point four or five gilaers, or you could attach an 367 00:18:32.440 --> 00:18:35.240 external one for longer range using special films, making thin, 368 00:18:35.480 --> 00:18:38.400 flexible tags possible. They're also working on batch assembly to. 369 00:18:38.319 --> 00:18:42.160 Cut costs, miniaturization driving everything right. So digging into the 370 00:18:42.200 --> 00:18:44.759 design of these passive tag ICs, what are the key 371 00:18:44.799 --> 00:18:45.599 building blocks? 372 00:18:45.759 --> 00:18:48.799 You need the rectifier obviously to get power, a voltage 373 00:18:48.839 --> 00:18:52.359 regulator for stable supply, a demodulator to get data from 374 00:18:52.359 --> 00:18:55.519 the reader, memory for the ID, a modulator to send 375 00:18:55.599 --> 00:18:56.799 data back, usually. 376 00:18:56.559 --> 00:18:59.400 Backscatter where flex the reader signal. 377 00:18:59.119 --> 00:19:02.319 Right, it doesn't transmit its own and a clock generator 378 00:19:02.319 --> 00:19:03.319 for timing the. 379 00:19:03.279 --> 00:19:07.119 Recifier seems critical for passive tags. How did they optimize that? 380 00:19:07.599 --> 00:19:10.680 They use things like the Dixon equation modified to account 381 00:19:10.720 --> 00:19:15.359 for real world losses, capacitance, diode characteristics, load current frequency. 382 00:19:15.880 --> 00:19:18.680 It's a complex balancing act choosing the right number of 383 00:19:18.720 --> 00:19:22.319 stages matching the antenna impedance to grab as much power. 384 00:19:22.119 --> 00:19:25.200 As possible, maximizing that energy scavenging. And once you have 385 00:19:25.240 --> 00:19:28.000 the raw power, you need stable voltage exactly. 386 00:19:28.359 --> 00:19:31.240 First, a DC limitter for basic protection against too much 387 00:19:31.319 --> 00:19:34.880 voltage could be simple diodes or fancier transistor circuits for 388 00:19:34.960 --> 00:19:38.799 tighter control, especially with low voltage processes. Then a fine 389 00:19:38.839 --> 00:19:42.000 regulator for the precise voltage needed by the circuits. 390 00:19:41.640 --> 00:19:44.119 And some tags needed to work at multiple frequencies. 391 00:19:44.359 --> 00:19:47.799 Yeah, research into multi frequency rectifiers using coupling in switches 392 00:19:47.839 --> 00:19:51.559 to handle LFHF and UAHF. Even the clock oscillator had 393 00:19:51.559 --> 00:19:54.640 tight specs needed over one point nine middle hertz decent 394 00:19:54.720 --> 00:19:58.440 voltage swing, but using incredibly low power like under five 395 00:19:58.480 --> 00:19:59.119 hundred nano. 396 00:19:59.000 --> 00:20:02.920 Wa amazing engineer ring in such tiny, simple looking tags. Okay, 397 00:20:03.000 --> 00:20:06.759 last piece of the RFID puzzle. Printed electronics a totally 398 00:20:06.759 --> 00:20:09.000 different way to make tags, potentially. 399 00:20:08.519 --> 00:20:12.400 Game changing, using printing methods, often with organic materials, for 400 00:20:12.559 --> 00:20:16.880 low cost, large area, high volume production. Think beyond RFID two, 401 00:20:16.880 --> 00:20:19.160