============================================================== Guild: wafer.space Community Channel: ℹ️ - Information / general / The wafer.space gf180mcuD Run 1 ATBS ADC After: 2026-09-30 11:59 p.m. Before: 2026-11-01 12:00 a.m. ============================================================== [2026-10-09 8:57 a.m.] mole99 [2026-10-09 8:57 a.m.] mole99 Awesome 🥳 {Reactions} 🙌 [2026-10-09 9:20 p.m.] mithro_ What is the rough sample rate and resolution of the ADC? [2026-10-09 9:21 p.m.] mithro_ @Simi - Would also be great to link to the github repo and tape out information for the project. [2026-10-10 6:06 a.m.] simi150500 GitHub Repo: https://github.com/iic-jku/gf180mcu-jku-atbs-adc {Embed} https://github.com/iic-jku/gf180mcu-jku-atbs-adc GitHub - iic-jku/gf180mcu-jku-atbs-adc: This wafer.space tapeout in... This wafer.space tapeout includes the digital core of an ATBS ADC from the IICQC at JKU. - iic-jku/gf180mcu-jku-atbs-adc 2026-10_media/gf180mcu-jku-atbs-adc-860BC [2026-10-10 6:08 a.m.] simi150500 This TO is basically the digital core described in my master thesis here: https://epub.jku.at/obvulihs/content/titleinfo/12118473?query=Simon%20Dorrer {Embed} https://epub.jku.at/obvulihs/content/titleinfo/12118473?query=Simon%20Dorrer An open-source adaptive event-based ADC for bio-signal acquisition... 2026-10_media/12125281-81CAF [2026-10-10 9:47 p.m.] mithro_ @Simi - And rough sample rate and resolution of the ADC? [2026-10-10 10:06 p.m.] simi150500 This is the thing: for event-based ADC, you have different metrics. You can not really define a "general" sample rate and resolution since it depends on the input signal and its instantaneous slope. See here: https://www.researchgate.net/publication/396348843_Event-Based_ADCs_vs_Nyquist_ADCs_Rethinking_Performance_Metrics [2026-10-11 4:14 a.m.] mithro_ @Simi - Assume I know nothing about ADC design and just want to understand when I should be using this ADC in one or two sentences (verse when another option would be better?). Really, just looking for something high level enough that I can "pattern match" against people who might be doing projects that could potentially benefit from your design and I should send them your way. [2026-10-11 4:21 a.m.] mithro_ @Simi - IIRC @Mehdi's group has done some designs for ADCs on GF180MCU. Would love to know the pros and cons of each groups choices! [2026-10-11 12:58 p.m.] enited Looks great @Simi ! I am curious if you are posting the silicon results back to the repo? [2026-10-11 2:41 p.m.] simi150500 Yes, sure. 🙂 [2026-10-11 2:46 p.m.] simi150500 Summary from my thesis: "The most commonly used ADCs are so-called Nyquist ADCs, which operate at a fixed sampling frequency and thus output equidistant samples [4,5]. However, certain signal groups exhibit significant variations only within short time intervals. This is particularly relevant in bio-signal acquisition and processing, where sparse signals are observed in electrocardiograms (ECGs) and electroencephalograms (EEGs). According to the Nyquist-Shannon sampling theorem, the sampling rate of a Nyquist ADC is determined by the highest frequency component during these transient changes [6]. This results in an unnecessarily excessive number of samples in minimal signal variation time intervals, leading to higher power consumption in data processing and transmission [7]. However, optimizing data handling and energy consumption is crucial for batterypowered, wireless, and wearable devices, such as fitness trackers, medical sensors, and Internet of Things (IoT) devices [8]. Event-based ADCs offer a promising solution by generating data points only when the signal exceeds a predefined voltage threshold [9–11]. The reduced data samples are strongly tied to savings in energy consumption, storage resources and computational effort [12]. In [13, Fig. 13], it is demonstrated that event-based ADCs consume a lot less power than the Nyquist ADC-based counterpart. As demonstrated in this work, adaptively adjusting these voltage thresholds further enhances efficiency [14]." {Reactions} 🙌 [2026-10-11 11:33 p.m.] mithro_ @Simi - I would summaries this as: @Simi has an ADC which optimizes for low power things and useful for battery power setups which have "human scale" O(1 kHz --> 100kHz) type requirements (wearables, biotracking, etc). Works best when signals have clear "events" in them. Is that correct? [2026-10-12 8:00 a.m.] simi150500 Yes, exactly. Signals that have small voltage changes most of the time but have short time periods with huge voltage ranges are perfectly suited for these event-based ADCs. And ECG signals are a perfect example for such "sparse" signals. 🙂 [2026-10-12 8:03 a.m.] simi150500 These figures show the functionality pretty well. During the QRS complex, you have many more data points. If the threshold is even adaptive, then you can further reduce the number of samples without loss of information when using correct reconstruction algorithms. 🙂 {Attachments} 2026-10_media/image-67A40.png 2026-10_media/image-327E4.png [2026-10-12 8:06 a.m.] simi150500 However, the huge "problem" of this ADC is now that the data samples are non-equidistant. So things like FFTs are not possible, and therefore showing the frequency domain of the reconstructed signal is not directly possible without interpolation. Meaning, metrics like ENOB and SNDR are, strictly speaking, not really possible. This is also the topic of my paper above. [2026-10-12 10:18 p.m.] mithro_ @Simi - Very cool! Great to hear that it is working. {Reactions} 🙌 [2026-10-15 2:54 a.m.] namibj Ehhh, it's not _that_ bad.... [2026-10-15 6:01 a.m.] simi150500 If you write a paper and want to compare it to state-of-the-art this is bad. Everyone is using different signals. 😆 [2026-10-15 6:03 a.m.] simi150500 And comparing to Nyquist ADCs is even worse since a sine wave input is not a good reference signal for event-based ADCs to show its performance. And then you cannot really say an ENOB etc. [2026-10-15 6:25 a.m.] namibj ehhh while you're limited by patterns fitting your buffers (unless your architecture is somehow still able to sustain maximum front end sample rate continuosly on signals demanding just that), ENOB is on it's own quite a local measure. You do basically still just take an ideal expected waveform and compare with the ADC's output bitstream, and just look at how much entropy there is. From looking at your block diagram it seems you're built like a hybrid between SAR and discrete-time-delta-sigma, leaning towards the SAR, and focusing on what looks like getting more samples if the signal changes too fast for the LSBs to settle. Like, a SAR who's sample-and-hold input block was taken away, kinda. [2026-10-15 2:27 p.m.] simi150500 "hybrid between SAR and discrete-time-delta-sigma" but this would deliver equidistant samples, right? [2026-10-15 6:02 p.m.] namibj I mean, you're kind of doing a dual-sided SAR that "compresses" no-change/no-trip time quanta away.... You have the lack of input sample and hold from a discrete time delta sigma, and the comparator second input fed from a DAC of a SAR. By having two comparators, one for lower and one for upper bound, you can drive the system such that you can keep high responsivity to transients that blow outside the band between upper and lower threshold, without tripping the digital machinery and DACs and whatever resolving/positive-feedback cascade the comparator needs, if the signal stays within the band that you keep as narrow as you _want_ it to be per the requests put onto your digital control logic that decides what DAC values to feed which comparator at what time. {Reactions} 👍 ============================================================== Exported 24 message(s) ==============================================================