RE: [SI-LIST] : predriver control schemes (was Some Semiconductor

West, Todd ([email protected])
Thu, 15 Jul 1999 13:53:29 -0700

> Slow-mode USB drivers, with a minimum edge
> time of 90ns,

Nitpick: actually, it's 75ns, 10%-90%.

> still aren't easy to guarantee full settling in
> 667ns. No big deal -- that's what makes this job
> fun.

Mind elaborating? Most of the low-speed drivers I've worked with use a
feedback loop to control the predriver's gate voltage and have excellent
convergence to full settling in 200ns or so. That's not to say, of course,
that there aren't plenty of folks who get it wrong. ;~)

Todd

-----Original Message-----
From: D. C. Sessions [mailto:[email protected]]
Sent: Thursday, July 15, 1999 12:40 PM
To: [email protected]
Subject: Re: [SI-LIST] : Some Semiconductors are Unnecessarily Fast

"Matt (boomer) Russell" wrote:
>
> hi,
>
> The problem with the current controlled predriver approach is that
> to achieve a constant current the gate voltage of the output
> device needs to be almost constant (very slow) until it goes
> linear vds<vgs-vt.

Not really. You use a large device with low gate bias to provide
the predriver current (remember, this is PREdriver current, not
final drive current) and switch it with the usual (relatively)
small devices.

> For a high drive cell with a low cap load (you want the bus
> to look capacitive [that is the whole point of the edge rate
> control]) That may be almost all the way across the output swing.

I think you're confusing output current with predriver current.
You do NOT want the bus to look capacitive; you don't CARE. What
you do want is to have the gates of the final drive devices look
capacitive, and that's easy because that's what CMOS does whether
we like it or not. Then a steady predriver current (Idsat) from
the predriver gives a steady dv/dt on the gate of the final drive
device, which (times Gm) gives a steady di/dt at the pin.

> This means that your predriver current needs to be very weak
> which means that the max frequency the IO can support will be
> very slow if you want the predriver to go full swing (otherwise
> you start to see data-dependant jitter).

Because the output switch occurs in a relatively small part of
the gate charging curve, controlled edge rates ALWAYS impose a
tradeoff with DJ. Slow-mode USB drivers, with a minimum edge
time of 90ns, still aren't easy to guarantee full settling in
667ns. No big deal -- that's what makes this job fun.

> plus you are assuming a process independant current source
> for the predriver....

This was in the context of an external programming resistor, so
you're absolutely correct. The same results could be obtained
by tapping a bias current off of the sysclock PLL, though, and
that would do a nice job of keeping the edge time proportional
to the cycle time across PVT.

> "D. C. Sessions" wrote:
> >
> > Mike Degerstrom wrote:
> > >
> > > Raymond,
> > >
> > > I'd also like to know more about the edge-rate control with use of
> > > a series resistor as has been mentioned is used with the Analog
> > > Devices 21020. I have seen mention of parts that use a reference
> > > resistor to control output impedance, i.e., drive strengths such
> > > as in the HP article by Esch, and Manley. I also noticed that
> > > IBM sells SRAMs with programmable impedance control drivers.
> >
> > In principal, you can control edge rates by setting the predriver
> > currents. Output edge rate is mainly dictated by the risetime
> > of the final driver devices' gate voltages, and that in turn is
> > dictated by the Ids of the predriver transistors. If you provided
> > the predriver with a bias voltage to a two-transistor stack (or
> > three with tristate) such that the switching transistors just
> > gated the current from the bias transistor the result would be
> > a controlled dv/dt at the final drive gate and indirectly a
> > controlled edge rate.
> >
> > Since the gate capacitance and transconductance of the final devices
> > are roughly proportional with process, this approach would give
> > a reasonably stable output edge rate. Of course there would be
> > timing impact, and if you controlled the turnoff transistors you'd
> > have crowbar current. (If you don't control them, then the turnoff
> > transition could get VERY abrupt, making the net output a bit odd.)
> >
> > So it *could* be done, albeit with tradeoffs in power, noise, and
> > layout complexity.
> >
> > --
> > D. C. Sessions
> > [email protected]
> >
> > **** To unsubscribe from si-list: send e-mail to
[email protected]. In the BODY of message put: UNSUBSCRIBE
si-list, for more help, put HELP. si-list archives are accessible at
http://www.qsl.net/wb6tpu/si-list ****

-- 
D. C. Sessions
[email protected]

**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****

**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****