Re: 1000V input, 5V output, 5W, SMPS reference design needed

From: Winfield Hill (Winfield_member_at_newsguy.com)
Date: 07/29/04 

Date: 29 Jul 2004 06:25:10 -0700

Alexander Odyniec wrote...
>
> My needs and concerns are similar to those described by Doug McNutt 7 years
> ago (below) but less stringent: I don't need a wide input range; just the
> fixed 480 V rms; non-isolated OK. However, I would prefer a solution that
> does not generate alot of EMI. Can anyone recommend a good design, please?
>
> From: dmcnutt@macnauchtan.com (Douglas P. McNutt)
> Subject: micropower 1000V MOSFET power
> Date: 1997/12/01
> Message-ID: <dmcnutt.1231365044A@news-2.sni.net>#1/1
> Organization: The MacNauchtan Lab
> Newsgroups: sci.electronics.design
>
> Mr. Hill's 1 kV design is of great interest to me because I have encountered
> the same design in the real (commercial) world and solved it in much the
> same way. I do have some remaining problems though which I hereby put up for
> comments. ...

 McNutt is referring to one of a series of posts I made in Nov 77, playing
 at my computer with various 1000V circuits, evidently while not having
 much else to do in the early AM hours (at work I had been designing new
 versions of 1kV amplifiers, using the same parts as in these posts).
http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&threadm=65njc2%24snu%40fridge.shore.net

 Here's a copy of one relevant post. You'll notice I ran this switcher
 very slowly to keep the standing power consumption down, and I sneakily
 avoided showing the controller portion of the circuit, its power source,
 or any startup circuitry. Nor did I discuss transformer design. :>)

 My favorite mtp1n100 and 1n120 low-capacitance high-voltage MOSFETs
 have long been discontinued by Motorola & ON Semi (sad, but I keep a
 small stock for critical instruments). If anyone wants a copy of the
 1n100 data***, I can send one by email. Anyone seriously interested
 in evaluating the design ideas can redo the calculations with a larger
 1.2kV FET, such as the IXTP3N120, http://www.ixys.com/98844.pdf

From: hill@rowland.org (Winfield Hill)
Subject: Re: Micropower 1000V power MOSFET dc-dc converter
Date: 1997/11/28
Message-ID: <65njc2$snu@fridge.shore.net>
References: <64f36s$2hl@fridge.shore.net> <64hou6$ib0@fridge.shore.net>
<64s7b8$5qb@fridge.shore.net>
Organization: Rowland Institute for Science
Newsgroups: sci.electronics.design

Winfield Hill at hill##rowland##org says...
> Winfield Hill wrote...
>> Winfield Hill said...
>> ... Writing about the nanopower and micropower linear operation of
>> power MOSFETs, continued ...
>>
>>> We'll start with Motorola's smallest 1kV power MOSFET, the MTP1N100E.
>>> The 1N100 is a 1A 1000V FET with R_DS(on) = 6.7 ohms (typ) and comes
>>> in a 75W TO-220 package. It's not a small transistor, and has an
>>> input capacitance Ciss = 587pF. But it's the smallest 1kV FET I've
>>> been able to find, does anyone know of one smaller?
>
> I received email suggesting Supertex and their line of high-voltage
> power MOSFETs, which they offer in TO-92 through TO-220 packages. Yes,
> we shouldn't ignore the extensive line they have, with many different
> die sizes, both P and N types, and even high-voltage depletion-mode MOS
> FETs, which can be very useful in micro and nano-power circuitry. But,
> as far as I can tell, their FET product line doesn't go above 600V.

 In this thread we've discussed several nanopower and micropower 1kV
 circuits, using power FETs.

 Here's an unusual micropower DC-DC converter; it works with a 1000 VDC
 input! Once again, I'm seeking the smallest 1kV MOSFET available, and
 settle on the Motorola 1N100, the TO-220 case power part detailed above.
 You may remember that typical selected FETs have only 2nA leakage at 1kV
 (and can operate to 1200V), so we know the converter's quiescent power
 consumption will be at least 2uW at 1kV.

 Here I'll spell out some of the design details for a converter that
 generates 15V and 5V outputs, has an average input-operating current
 of about 10nA with no load, an efficiency of 70 to 75% for loads above
 50uW, and a power-output capability of about 3 watts, running at 25kHz.
 I've chosen a flyback-transformer configuration, with a turns ratio to
 create a +100V primary flyback (i.e. the voltage on the FET will be
 1.1kV plus the snubbed leakage-inductance spike).

  + 1000V 200V piv
   IN ---+-----+----------, ,------>|-----+---- 15V
         | | O || O 36t 150uF
         | | O || O |
         | snubber O || '- gnd gnd
         | | O ||
         | | O || ,----->|---+---+--- 5V
       0.1uF | 225t O || O 12t 470uF |
       1.5 kV | O || '----------+------- gnd
         | '------+---' core |
         | | A_L = 1000 | 5V current drain
         | D nH/N^2 | must be more than
         | -- G | 3x the 15V current
         | S MTP feedback
   1kV | | 1N100E
   rtn --+-- 6.8 -----'
                      

 We'll start the design by considering the energy lost in a single cycle
 of the converter. This will allow us to scale the parameters so the
 conveyed energy is much greater than the wasted energy. Our problem
 here is the total capacitance of the transformer, power FET and snubber
 diode, which is charged from the 1kV power source and discharged into
 ground, once each cycle. Interested readers will want to follow along
 with a 1N100 data *** (in the Motorola TMOS book, DL135/D rev 6 page
 4-678, or http://www.mot-sps.com/books/dl135/pdf/mtp1n100erev2x.pdf).

 It's not easy to determine the charge in the FET's output capacitance
 from the 1N100 data ***, because of its very nonlinear nature. For
 example, Coss = 18pF at 400V, but 60pF at 25V and climbs to 600pF when
 saturated at 1V. If we assumed a constant 40pF, the wasted charge would
 be q=CV = 40nC for each 1kV excursion (or 12.8nC for 320V).

 Motorola does give a detailed log-log plot for Crss (the gate-drain
 capacitance), and Coss (the output drain-source capacitance, plus Crss).
 From this (fig 7b) we see Coss vs Vds tracks and is always about 6*Crss.
 They also give detailed charge data for the gate (fig 8), which careful
 examination shows about 2.3nC for a 400V to 80V Vds excursion (during
 which Vgs remains fixed at 5.5V). We get 6 * 2.3 = 13.8nC, or close to
 our 12.8nC estimate above for 40pF. So, can we also believe our 40nC
 estimate for the full 1000V swing (i.e. simply multiply the charge by
 1000/320)? Examining the gate charge graph for the region Vds = 400V
 down to 0V (afterwhich the gate drive causes the gate voltage to begin
 rising again), we see the gate capacitance has consumed about 7.5nC,
 implying 6* 7.5 = 45nC for the drain-to-source. Adding 18pF * 600V =
 11nC to account for the 1kV to 400V portion, we get 56nC total.
 This is more realistic than our 40nC guess.

 Now perhaps we can calculate the dynamic switching loss. We'll add 7%
 for the snubber diode (a MUR1100E is 3.5pF at 25V), or 4nC, and we'll
 budget 25pF for the windings and transformed secondary capacitances,
 adding 25nC. We get 56 + 4 + 25 = 85nC.

 Now we can calculate an energy loss of 1/2 CV^2 = 1/2 qV = 43 uJ per
 cycle. Let's say we want 75% maximum flyback-stage efficiency. With
 a little manipulation we write Eo = Eloss (n/1-n) where n=0.75, the
 efficiency, and get about Eo = 125uJ per cycle. It's interesting to
 note that operating with a 5uW load, only one cycle is needed every
 125/5 = 25 sec. At a more reasonable 5mW load, the converter runs at
 40 cycles/sec.

 The high per-cycle energy also means the output storage capacitor will
 be larger than usual, say 470uF for the 5V supply, and a selected
 low-leakage part will be appropriate. A small Rubycon 160V electrolytic
 I have in the shop measured 0.2uA leakage at 5V (this is 5/0.2 = 25M
 leakage resistance, or 12,500 M-ohm-uF, much better than official specs).
 Quite a few 470uF parts from various manufacturers measured around 0.3uA
 after soaking a few minutes. So we can estimate 5V*0.3uA = 1.5uW loss
 for the 5V storage capacitors. We'll assign another 1.5uW to the 15V
 output capacitor, and a massive 2.5uW to the voltage-sensing and control
 circuitry (sticking to that budget is another story).

 Finally, let's assume 200000-M-ohm-uF 0.1uF input capacitor (0.5uW at
 1kV), so we can now calculate our overall quiescent power consumption,
  2 + 0.5 + (1.5 + 1.5 + 2.5)/0.75 = 10uW.

 Since we want an energy of 125uJ per cycle, we can calculate L = 2E/I^2
 = 2*125/0.07^2 = 50mH for a 70mA peak transformer primary current. With
 25mH charging inductance, we get t = I L / V = 3.5us charging time, and
 since the primary discharge voltage is 100V, we'll get a 35us secondary
 discharge time. If the converter is operated above 1/38.5us = 26kHz,
 or 3.25W power level, it'll have a continuous inductor current, which
 is just fine.

 OK, folks, that's it for this installment. 

-- 
Winfield Hill      
Rowland Institute for Science
Cambridge, MA 02142
 Thanks,
    - Win
 (email: use hill_at_rowland-dot-org for now)