LM2901B: Negative output spikes with low current pullup

The LM2901B does not have an internal compensation capacitor.

This device has an open-drain output and cannot generate voltages below GND.

The decoupling capacitor (C58) does not have good (short, low-impedance) traces to U8's power rails. It should have been routed with priority.

Please show an oscilloscope trace with the voltages at pins 7, 8, and 14.

Hello Rusty,

Also check waveforms for pin 12 GND and pin 1 VCC for this noise signal.

Hi Rusty,

1) No

2) Wrong internal diagram - see Clemens schematic.

3) No - fix the problem...

Also - the layout shows pins 11 and 12 connected together along with the pad...is this the correct schematic/layout?

Is there a reason the LED bias resistors are soooo precise (680//2k)? Seems a little extreme..

The output can only pull to GND, so there is no way the comparator output can pull below the GND pin.

The ONLY way that the output could pull below ground is that some pin needs to be pulled WAYY below ground to turn on internal parasitic devices. When parasitics get involved, boundaries between channels fall and one channel being abused can affect another channels through the substrate.

Do the negative spikes correspond to any other switching in the system?

Also - try increasing the value of C58 (supply bypass) - to maybe 100uF. Both the comparator, references and LED's are powered off the same "soft" supply, and the LED's turning on could cause a dip in the supply, causing feedback.

Thank you guys for the quick responses.

I'm convinced this is inherent behavior of the LM2901B and not a schematic or layout issue.  A simple pspice simulation of the LM2901B output stage shows negative glitches on the output with high pullup resistance.

Please see the attached PDF for responses to your comments, an explanation of how I think the negative glitches are generated, and a simple pspice simulation showing the negative glitches.

I'm also milling a test coupon with only the LM2901B on it using an optimum layout.  Should have that tomorrow and will post waveforms.

LM2901 comparator glitch 02.pdf

The comparator's output is digital, so current spikes can happen when switching.

Please show an oscilloscope trace of pins 7, 8, and 14. The repeated negative spikes appear to indicate that the hysteresis does not work correctly.

This comparator is analog.  The output stage steers a current source to or from the final bipolar output transistor.  Shoot-through, crossover, current spikes, and other similar things seen in digital do not happen here.

The negative pulses are due to changes in the LM2901B output stage compared to legacy versions that do not exhibit this behavior.  Negative spikes appear when the LM2901B output load current (pullup current) drops below about 1.2 mA.

These negative pulses:
1)  Change the trip point voltage and delay the trip point when hysteresis is used due to initially providing negative feedback until the output goes positive and provides the desired positive feedback.

2)  Exceed the Vin ab max rating of the datasheet.  The negative pulses exceed -500 mV and lasts hundreds of nanoseconds.

3)  Also occurs when an output pin is left floating and that particular output is switched from low to high.

I built a simplified test board.  The schematic, layout, photos, and scope traces (including VCC, Vout, +IN, and -IN) are attached in the PDF below.  Let me know if you would like more data.

LM2901 comparator glitch 03.pdf

Thank you for creating the test board.

I've noticed that there is no good current path to ground for pins 8, 9, and 11, i.e., the input bias currents have to flow through high-valued resistors and will create a measurable voltage drop; and the path for pins 8 and 9 changes when the output switches. But this should result in additional positive feedback, so is probably not related with your problem.

In any case, the LM2901B was never designed to handle currents as small as you're using, so it would be plausible that such a situation was never tested.

For this application, I would use CMOS comparators (e.g., TLV1824, or better TLV1814 with push/pull outputs).

Rusty,

I can look at this later this week in lab. It is an oscillation loop that generates transients. 

Thanks, again for the responses.

Clemens: The LM2901B has buffered inputs (Q1 and Q4 in your schematic) to provide constant input current. The problem exists even when the input resistors are very low value or even no resistors at all (i.e. square wave pulse input).

The issue occurs when the output sink current is less than roughly 1.2 mA and is independent of all other circuit or layout factors. Datasheet Figures 7-19 though 7-22 show curves for sinking current down to 10 uA. So, it was certainly DC tested below 1.2 mA.

TLV1824 and TLV1814 are 10x the cost. The legacy LM2901/A does not have this issue but is also not available in the small WQFN package.

Ron: This is NOT an oscillation loop. The negative transient on the output occurs any time the output transitions from low to high when the output sink current is less than about 1.2 mA. You can drive the inputs with a square wave from a pulse generator and the negative transient still occurs (but just once per cycle). The oscillations in my latest pdf are due to the negative pulse initially creating negative instead of positive feedback.  In that particular circuit the negative pulse creates the oscillation - not the other way around.  Please see my second pdf above (LM2901 comparator glitch 02.pdf). This explains the likely physics of why the comparator output stage is generating a negative voltage and includes a crude pspice simulation of the output stage that shows this negative transient generation.

RUSTY BOUDREAUX said:

Ron: This is NOT an oscillation loop.

It has to be a loop. It oscillates and generates voltages beyond rails. I didn't say where this loop is located. It could be completely inside LM2901B, I have not looked for it yet. I will look, I will find. 

In the pdf I explain the mechanism for negative voltage beyond the rails that does not require oscillation, internal or otherwise.



Short version:  When the output is low, Q8 collector is near GND and Q8 base is +VBE above GND.  To allow the output to go high Q13 turns on to drive Q8 base near GND.  However, the Q8 collector-base capacitance takes time to discharge.  So when Q8 base goes from +VBE to GND the Q8 collector goes from GND to -VBE and the output is negative until this capacitance is discharged.  This is easily simulated as shown in my pdf.

Only when the collector pullup current exceeds the Q8 base current (Q13 collector - Q14 collector) does the output remain positive during the low to high transition.

Something has changed in the output stage of the new device.  Q13/Q8 geometry, Q13 speed/current, or some method of saturation control (Baker clamp).  The legacy LM2901 and LM2901A do not exhibit this behavior.

Rusty,

I tested these in the lab today. I'll share results tomorrow.

Rusty,

These spikes are a failure of the VOH drive. I highlighted the first attempt for VOH (Q8B going low) that was unsuccessful.

It VOH did succeed by going high, then hysteresis would hold that high preventing output chatter. 

Below is device schematic with Q8 collector base capacitance highlighted in red. Q8 is by far the largest transistor in LM2901

1) LM2901 has no output sourcing capability at all. The application has to take over VOH drive role, no matter what it takes.

2) LM2901 is a intended to be a forward only linear amplifier with no frequency compensation. So slow input voltage leads to output chatter.

3) While there are no intentional loops, that shared "internal bias bus" does allow a small amount (greater than zero) of loop feedback.  

4) External hysteresis requires large output swing on first attempt to avoid output chatter.  

5) Q8B drive going low quickly (B devices are quicker and have high DC gain) will capacitively pass up to 1 VBE of negative voltage to high Z output. 

The failed VOH attempts happen at regular intervals dependent on the input voltage difference. This effects seems to be bigger than random noise generated output chatter.  So unintentional feedback dominates. Point #3  

RUSTY BOUDREAUX said:

Something has changed in the output stage of the new device.

Yes, Q13 and Q14 are added for more gain. Most transistors are smaller, less parasitic capacitance , faster.  

RUSTY BOUDREAUX said:

The legacy LM2901 and LM2901A do not exhibit this behavior.

That statement is incorrect. Legacy LM2901 has this behavior to a lesser extent. I will show this in next post.

Rusty,

I ran out of time for today. Lab data and test setup tomorrow. 

Thank you, Ron.

External hysteresis may be the main cause of the failed VOH attempts at regular intervals.  When the 1st VOH attempt goes negative the external hysteresis resistor pulls +IN down causing a switch back to VOL.  As +IN slowly rises the cycle repeats until the +IN dip caused by the external hysteresis resistor doesn't cause a switch back to VOL.

The intent of external hysteresis resistor is to provide clean switching for slow input signals.  However, this negative output pulse causes the opposite - the external hysteresis resistor increases chatter.  Increasing external hysteresis actually increases chatter since it feeds back a larger negative dip on +IN.

The only solution I've found is increasing output sink current above roughly 1.2 mA which prevents the negative VOH swing.  A small cap from +IN to -IN can also help as it filters the brief negative pulse before the much larger VOH swing provides the expected hysteresis.  But a cap here is usually not recommended since it slows the hysteresis.  The trick is to filter the small negative pulse but not the larger VOH swing.  A diode from Q8 base to collector (a Baker clamp) would prevent the stored charge in Q8 that causes the negative spike but obviously that's not externally accessible.

Rusty,

Here is the open loop lab setup. Vcc = 10V, Vpu = 0.1V, IN+ sweep is 1mV wide over about 5 seconds. A very slow sweep and a very weak pull up current.

Scope 1M matters here, so I included it in schematic.  Power on (VCC) when the input sweep occurs is blue.  

Here is the legacy 2901 waveform.  I highlighted the negative voltage output chatter. After that comes usual output chatter (present in push pull comparators) . 

Below is the LM2901B waveform. The width (time and input voltage) of the chatter is less (good).

The amplitude of the negative chatter is much greater (not good).