(New, Pre-production Preview, November 2021) ISOUSB211 - Low emissions, high/full/low speed isolated USB repeater

This is absolutely phenomenal. Texas Instruments just made a galvanic isolation ASIC for 480 Mbps (not 12 Mbps) USB 2.0 that you can actually buy. Finally this problem is solved at last.

· · Web · 3 · 6 · 5

I ordered the pre-production prototype version of TI's high-speed USB isolation chips as soon as I got the news. My order has just shipped today! :blobcataww: Time to design a development board.

> isolation barrier capacitance, input to output: 1 pF
wow, that's some serious galvanic isolation.

Mirror, mirror on the wall, who has the highest breakdown voltage of them all?

USB Killer ordered, I'm going to unironically use it as a test device on my 1000 volts port isolator...

BTW, it was claimed that some MacBook models are immune to the USB Killer because the ports have optocouplers but I failed to see any evidence. This is a highly unlikely claim. I wonder what could be the real reason... Pre-enumeration current limiting? CMOS data switch?

Single-chip optocouplers above 100 Mbps for mass markets simply don't exist at all, nearly all vendors use capacitive or RF isolation at higher data rates. And even with an isolator, USB 2.0 HS is bidirectional, you need a controller with state machines to follow the protocol and do bus arbitration, challenging enough that no company ever made such a chip for the open market until Texas Instruments in 2021.

What makes it even more unlikely is that, even when the port is optoisolated, the USB Killer would still kill that USB port *regardless* of how the port is implemented. But according to the alleged MacBook test report, "unable to discharge [high current from the capacitor]", which doesn't make any sense. The only two reasonable explanations are either the USB port's electrical behavior somehow depends on whether a device is detected, or that the port has an unusually robust input protection design, e.g. Bourns's TBU High-Speed Protectors can electronically disconnect the signal line within 1 microsecond after a high-voltage transient, and can itself withstand hundreds of volts. Both solutions sound unusual, but at least possible, unlike the optocoupler claim.

* State-of-the-art microprocessor SoC built using the 7 nm node, one of the most complex process ever invented by humans.
* A capacitor charged to 300 volts.

Who would win?

The last overvoltage protector candidate has arrived. Time to send 300 volts to each part and see which one blows up first...

USB Killer waveform captured - 4 kHz inverse sawtooth wave, 622 ns rise time, 172 volts AC peak-to-peak. The sawtooth wave itself is riding on a 138 volts DC bias, giving a total peak voltage of 310 volts AC + DC. The DC bias decreases as the stored energy goes down, the pulse stops after 20 seconds.

Unlike the original version, this particular variant uses a positive, not negative surge, and also adds a huge DC bias to the pulse. Its designer must have spent much time on improving its destructiveness. And curiously, it attacks only D-, not D+ data line, not sure if it's a manufacturing defect, or an intentional attempt to concentrate energy on a single line.

I put a 50 Ω load on the USB killer. At first it kept pulsing continuously for 30 seconds or so and stopped. Now it only generates one or two pulses into 50 Ω each time. Also, when the load is removed, the total pulsing time to a high-impedance load has also degraded from 20 seconds to 10 seconds...

Has this USB Killer just killed itself for good? :blobcatgiggle:

Curiously, each time the USB killer discharges, there's a loud "BANG" sound. A Gas Discharge Tube is used as the trigger it seems. It's kind of ironic, a GDT is normally used to protect a device from high voltage destruction, not to create it...

After seeing one myself, I think USB killer really sucks as a test device. If you connect it a 50 Ω load, it kills itself first because the energy dissipated is too high for the capacitor and switch in such an undersized device. Many are basically a disposible device - the MOSFET dies immediately after the victim USB port is shorted to ground as well... But this alone is not a problem, you can just limit the repetition rate to 1 pulse per minute with a manual trigger switch, which is obviously also not provided by most variants... Ideally it also needs adjustable pulse amplitude and polarities. Unfortunately the only practical application of a USB Killer in its current form is sabotaging, and even in that application it's not guaranteed to survive...

Conclusion: The USB Killer is just a meme. Time to shop for some DC-DC converters and HV transistors to make my own HV pulser...

> testing the pulse-energy endurance of electronic components such as resistors and transient voltage suppressors.

I think I don't really need to design this HV pulser. Just realized the X Chapter in The Art of Electronics already has exactly what I was looking for.

BTW, the USB Killer appears to have successfully destroyed my 20 dB RF attenuator during the test. :blobcatgiggle:

shorted to ground, now it's completely dead.

The original plan - let the USB Killer send its energy straight into a 50 Ω oscilloscope input through a big RF attenuator - is flawed. No reasonable RF attenuator can sustain such an energetic pulse.

New plan: Probe the voltage via a 450 Ω or 4950 Ω series resistor, this forms a 10x or 100x transmission line probe when combined with a 50 Ω input with good signal integrity. Also, the load should be installed on the PCB, using the instrumentation itself as a test load is a bad idea...

Success. The 100x transmission line probe (5.1 kΩ in series with 50 Ω oscilloscope input) is working really well, so far I haven't blow up my oscilloscope terminator yet. I'd say it's the go-to choice for high-voltage pulse measurements if you don't need higher input impedance.

More USB killer waveforms captured, unlike the last test (see thread) with an open circuit, this time it's connected to a 50 Ω load.

It keeps pulsing once it's connected, peak voltage is typically 140 V, but can be as high as 200 V. Pulse width is about 100 μs to 250 μs. Repetition rate is about 27 pulses per second. The pulse characteristics are poorly controlled, I suspect it's triggered by the arcing inside a gas discharge tube, typically used for surge protection (the irony...).

Like the last unit, this (identical) one also damages itself after continuously pulsing for a minute or so. It can no longer generate continuous pulses.

Received two 100 volts electric shocks while testing this, Electroboom moment. Note to self: Do *not* touch the output capacitor bank without discharging... Even after the output went to zero, it only means the voltage is too low for the switch to turn on, but the capacitors themselves are still at 100 volts!

Show newer

Board almost finished after yet another month of procrastination and doing nothing.

I can't wait to see some spectacular fireworks from a smoke test.

Hmmmm. So I need 20 millimeters of creepage distance for 1000 volts per IEC60950. That's kind of... Creepy.

So this is what I need to do for the required creepage distance. I think the whole board is just gonna crack into two pieces.

Anyways, I'm now pretty sure that those cheap $20 USB isolation dongles are definitely not designed for continuously working at 1000 V. It's just impossible when the required clearance and creepage are literally bigger than your entire dongle! Of course, unless it's only for functional isolation, for which there's no distance requirement at all.

On second thought, the isolated DCDC converter on my board only has functional isolation, and this already seriously limited the working voltage of the entire system. It's kind of pointless to design the rest of board for reinforced isolation per IEC 60950. I guess it's time to remove the slots.

After closer inspection, it turned out that the astonishing 1500 volts (AC) & 2100 volts (DC) "Maximum Isolation Working Voltage" boasted by Texas Instruments is only an industry standard for digital isolation *components*, while authentic, it's not for complete *systems*.

When you actually use the system-level safety standards used by national regulators like IEC 60950 and IEC 60664, you'll find in a real application, the 8 mm creepage distance at the chip package limits the working voltage to no more than 800 volts AC (Material Group 1, Pollution Degree 2, Reinforced Insulation). So much for the marketing.

> The large print giveth and the small print taketh away. -Tom Waits

> The nice thing about standards is that you have so many to choose from; furthermore, if you do not like any of them, you can just wait for next year's model. - Andrew S. Tanenbaum

There are more gotchas.

> Clearance: 8 mm
> Care must be taken during board design so that the mounting pads of the isolator on the printed-circuit board (PCB) do not reduce creepage and clearance.

Then the 8 mm clearance is immediately violated, officially, in the recommended footprint at the end of the datasheet.

Another pitfall is the IPC-2221 standard in most PCB calculators. Its clearance requirement can get ridiculous for some use cases because of the stepwise definition, yet insufficient for other uses cases (reinforced insulation).

"Friendship ended with IPC-2221, now IEC 60950 is my best friend!"

UL sucks too. Their Recognized Component certificates are essentially useless. All they tell you is a certification exists, but without any information about the rated working conditions, which are critical to safety. A vendor can say its power supply works up to 1000 volts, and claims the power supply is also UL certified (without telling you it's certified only for 100 volts, the 1000 volts spec is only a functional rating and cannot be used for safety-critical applications).

The real information is in UL's Conditions of Acceptability, and it's often nowhere to be found. You can either try asking the vendor nicely and hope they don't ignore your request. Or pay (possibly thousands of dollars?) to purchase that information from UL.

PCB ordered. Added common-mode chokes at both sides of the isolated converter hopefully to suppress some noise across the barrier. Also removed the slot - the SIP DCDC module only has functional isolation, so increasing the creepage distance further is pointless...

Hopefully I won't accidentally kill myself later in the dielectric withstand voltage test by the 3000 volts HiPot tester. LIke many on the Fediverse, is a thing, but a HiPot tester is just a terrible way to die.

My USB 2.0 isolation board is working. The Texas Instruments chip ISOUSB211 works as advertised, it really is the first USB 2.0 high-speed (480 Mbps) galvanic isolation chip on the open market. I'll release all the design files tomorrow.

TI ISOUSB211 development board design is now public. This TI chip is the first ASIC capable of doing USB 2.0 High-Speed (480 Mbps, not 12 Mbps) galvanic isolation on the open market. As always, must be loved.

Been hoping for something like this for a while 😀

@xro I just spent a year working on a workaround for this problem. Now TI has basically made 90% of my work meaningless. But I'm glad it now exists.

@alexandra I'll eventually get a USB Killer for a destructive test.

@niconiconi obviously the USB Killer sets some kind of "evil bit" that the macbook detects! :D

@niconiconi sounds like something that can be answered by a tear down

@cinebox no need for tear down, you just need to read a leaked schematic from one of those repair site. But I haven't see any protection, not sure which model it was.

@niconiconi Since your computer is disposable in the situation, it doesn't matter the USB killer itself is disposable?

@river USB port isolation. There can be a 1000 volt difference between the primary and secondary side of the board. Useful for protection and noise suppression. TI's ISOUSB211 is the first chip on the open market capable of doing this for High-Speed (480 Mbps) USB.

@niconiconi ok, yes, I was just wondering about applications. I don't think I've ever had 1000V on a USB device.

@river The basic idea is to separate both sides into two independent power domains, at different electrical potentials without a common ground. So all the common-mode noise and transient currents are contained at one side without conducting back. Helpful for many noise problems, most common one is breaking a ground loop in an audio interface.

Also useful for hardware development, so the PC's USB port is always protected. The isolated ground also allows you to make a measurement between any two points with a grounded oscilloscope (without lifting the scope's ground and compromising safety).

A working voltage up to 1000 V is a bonus and useful for safety-critical industry systems to keep the operators and equipment safe when the secondary side has a catastrophic failure. This board is definitely not designed to do that.

@niconiconi I can see the oscilloscope thing being useful.

@niconiconi I think you probably want four anchors on each side, unless splitting in half is part of the design (-;

CAM engineer: Hey, you must double your payment. Your design has two boards.
Me: It's one board.
CAM engineer: We are not fools, stop abusing our prototype system!

@lynne my original motivation was embedded system testing with floating measurement and noise suppression for software-defined radio, but it should work for audio as well.

@niconiconi thanks for this, it literally just came up for me at work while looking at optocouplers

Luckily I don’t have an isolation requirement because the spec I’m following is vague as hell

Sign in to participate in the conversation

cybrespace: the social hub of the information superhighway jack in to the mastodon fediverse today and surf the dataflow through our cybrepunk, slightly glitchy web portal support us on patreon or liberapay!