RE: Discrete, mixed, and integrated multi-port solutions
Peter,
See my comments bellow.
Yair.
> -----Original Message-----
> From: Schwartz, Peter [SMTP:Peter.Schwartz@xxxxxxxxxx]
> Sent: ו, מרץ 23, 2001 4:22 AM
> To: Yair Darshan
> Cc: stds-802-3-pwrviamdi@xxxxxxxx
> Subject: Discrete, mixed, and integrated multi-port solutions
>
> Yair:
>
> Good questions, good observations. Thanks taking the time to reply. I'm
> going to answer you by the numbers; I hope I don't make a confusing chain
> for other people this way...
>
> 1. I consider that, with adequate protection, the MOSFET need be rated
> for 400mA ~ 500mA maximum, possibly less (as you know, that's an ongoing
> discussion).
[Yair Darshan] for 350mA continuous we will need 1A rating (stress,
reliability, peaks etc.)
> Implementing very rapid and rather sophisticated (e.g.,
> dual-level or dual-rate) overcurrent protection on an integrated MOSFET is
> not very hard these days. Good design practice also would thermally
> protect
> the MOSFET(s) integrated onto a monolithic device. This gives the
> safeties
> of slow overcurrent protection, short-circuit protection, and overpower
> (overtemp) protection. The added cost for these functions id not zero,
> but
> it gets lower every year, while remaining very elusive if not impractical
> on
> a "standard" discrete MOSFET. I agree that 80V is a minimum number for
> BV(dss).
> 2. Cost is a very different thing from either price or value. In some
> cases, an external discrete MOSFET will yield the lowest-price, or better
> still, highest-value solution. But in many cases, the design time, the
> board space, and/or the simple issue of qualifying and inventorying
> components from a number of different vendors makes the value - and
> therefore the price - of even a slightly more expensive fully integrated
> solution greater than any savings which might practically be realized in a
> quasi-discrete circuit implementation. By way of crude comparison, I can
> build a pretty good op amp with 5 transistors and 5 resistors, which comes
> out to less than 10 cents in component cost. But each of those 10 parts
> will cost 4 cents to place on a board, and the board area consumed will
> vastly exceed that of a 20 cent SOT-23 op-amp, which I can get from any of
> a
> number of companies. Which is the lower parts cost? Which is the better
> value?
[Yair Darshan] I agree. However this example is not suitable to our
case.
1. We are talking about 1 components vs two components.
2. pick and place cost for high quantities is 1 cent. thus it is not
relevant to our discussion.
3. We have to weight example as close as possible to our case : a
chip with mosfet /mosfets or a chip with external mosfets.
4. From layout and PCB routing point of view, and specially for a
12-24 port switch, the area near RJ45 is critical and needed for the data
hardware.
Routing of traces near RJ45 is important for getting data
integrity. Thus the multi port chips will be far from this RJ45 or will be
located on 2nd PCB floor.
It means that locating the chips in area A and the mosfets in
area B can help optimizing the routing of traces. this chips and mosfet are
working at low frequency/DC.
> 3. The functionality which an integrated device can subsume into
> itself, and the degree of MOSFET monitoring and protection which it allows
> at minimal extra effort, vastly exceeds anything which could be done in an
> 8-pin IC alone. That's one way of looking at it. A better analysis was
> given by my colleague Bruce Inn, who is more aware of the process details
> for IC fabrication than I am. I will say this, though: the digital
> revolution has made almost all of us into "mixed-signal" houses. The
> question is simply whether we take our principal value-added from analog
> or
> digital expertise. If the function requires digital, digital will be
> found
> that will do the function, at a reasonable cost (remember the first
> digital
> watches? Pulsar. Using Integrated Injection Logic).
[Yair Darshan]
I agree.
> 4. Having worked with the Taiwanese on board layout, I will
> respectfully take issue with you on this point. The Taiwanese do not LIKE
> surface-mount parts, as they would much rather be able to stand a device
> up
> on a single-sided PCB. And, if possible, to run at least one trace
> between
> each two leads of that device. The last refuge of the DIP package (Delta,
> Lite-On, AcBel, PhiHong). We are looking at a "skim-the-cream" industry
> today. But I anticipate that very soon the 802.3af standard will mostly
> be
> accommodated by Taiwanese and PRC vendors, or we will be back to wall
> warts.
> 5. Good layout always involves tradeoffs. This may in some cases force
> a quasi-discrete, or at least multi-chip solution, whether or not a single
> chip multi-port device would otherwise be desirable in the application.
> No
> contest on this point, my friend.
> 6. I hate to do this, but I refer back to my Statement #2. Cost,
> Price, and Value. Density, Inventory, Vendor Qualification, Design Cycle
> Time.
[Yair Darshan] I agree. But the bottom line is what will be the
final solution cost after integrating all the factors above.
Can we have numbers of similar cases to
compare?
> 7. One additional point: A device carrying on-board MOSFETs intended
> for the PSE market would have those MOSFETs purpose-built to fit the needs
> of the application, no more, no less. This would facilitate having the
> proper specs MOSFET specs out of the box for leakage at temperature,
> leakage
> at voltage, R(on), and thermal resistance. I probably missed some, but
> you
> get the idea.
[Yair Darshan] I agree.
> So there's my opinions. Always makes me think, when we chat.
>
> > Peter Schwartz
> > Applications Engineer
> > Micrel Semiconductor
> > Phone: 408.435.2460
> > FAX: 408.456.0490
> > peter.schwartz@xxxxxxxxxx <mailto:peter.schwartz@xxxxxxxxxx>
> [From Yair Darshan]
>
> Hi Peter,
>
> Thanks for your info, However now I am more confused., I got other info,
> and
> in order to be sure that we are talking about the same parameters I have
> some questions to you. See below.
> 1. Are you taking in account that the Mosfet need to be rated
> to 1A/80V with low enough Rdson that can meet our requirements?
> 2. Are you saying that integrating the Mosfet into the chip
> will cost less than using low voltage chip with external Mosfet?
> 3. Using low cost chip say 8pin if we are lucky with the above rated
> Mosfet will not drive us to larger package? Please advise?
> 4. This examples shows my point, they are TO220 package with 3-5 pins,
> and this is exactly my point you are not saving board space. It is about
> the
> same or worse, especially for multi-port.
> 5. In addition, better layout in some application is achieved if the
> Mosfet is located on Area A and the Control on Area B for controlling high
> currents etc., especially in a multi-port environment.
> 6. Finally, if you check your examples for high quantities you will see
> that Integrated solution will cost 1.2 - 1.5 times as much (Power
> Integration and others) as the discrete solution will cost:
> Controller + Mosfet = 0.4$ + 0.33$ = 0.73$
> which is less...
> Please let me know what is your opinion on the above comments.
> Thanks-
> Yair.
>
>
> > -----Original Message-----
> > From: Schwartz, Peter [SMTP:Peter.Schwartz@xxxxxxxxxx]
> <mailto:[SMTP:Peter.Schwartz@xxxxxxxxxx]>
> > Sent: March 22, 2001 10:11 PM
> > To: Yair Darshan
> > Cc: stds-802-3-pwrviamdi@xxxxxxxx
> <mailto:stds-802-3-pwrviamdi@xxxxxxxx>
> > Subject: PSE vs. PD power dissipation again
> >
> >
> > Yair, Dieter, and all:
> >
> > Speaking both as an engineer having started with thermionics and
> now
> > working
> > with integrated power MOSFETs, and as someone who has a direct
> exposure to
> > the semiconductor industry, I believe I can say the following with
> both
> > history and present knowledge to back me up:
> >
> > A discrete MOSFET in conjunction with a low-voltage controller
> circuit for
> > the PSE may be a lower parts-cost-per-port solution today, and for
> a year
> > or > two.
> > That is TBD. But eventually (sooner rather than later), the
> MOSFET
> > will be integrated, as board space and circuit complexity has a
> way of
> > becoming more expensive than silicon - or conversely, silicon just
> keeps
> > getting cheaper.
>
> > I suggest the examples of Power Integrations (the > "TopSwitch"
> family) and the similar ST Micro "Viper" family of one-chip > off-line
> power
> supplies.
> > Open any 'ATX'-type computer power supply. It > can't be done
> for the price, but by the miraculous alchemy our industry > works to turn
> sand into gold, it is.
> >
> > Conclusion: Multiple-port PSE switches with integrated MOSFETs
> will > become > an available, cost-effective, and viable alternative to >
> multiple-component-per-port implementations.
>
> > Thanks for letting me sound off here.
>
> > Peter Schwartz
> > Applications Engineer
> > Micrel Semiconductor
> > Phone: 408.435.2460
> > FAX: 408.456.0490
> > peter.schwartz@xxxxxxxxxx <mailto:peter.schwartz@xxxxxxxxxx>
> >