Friday, November 1, 2013

Krampus Wish List - Simple Electronics

This is part 3 of a series in response to http://thearchdruidreport.blogspot.com/2013/01/a-wish-list-for-krampus.html

I cover the simple electronic devices that could be made, subject to certain assumptions, in a "post-industrial" world where fossil fuels are depleted and humans must live sustainably.  In part 1 I have covered the production of silicon wafers, the starting point for a photovoltaics (PV) technology (further described in part 2) and for simple electronics devices, as described herein.

Motivation


Life today seems impossible without computers and smart phones, yet these are recent developments that depend on complex and expensive supply chains and support systems, that in turn depend on economies of scale, in turn depending on (cheap) fossil fuels to support a growth-oriented prosperity.  It's impossible to predict where things will end up in the economic trade-offs, but without the money push to buy the latest new features, some small number of semiconductor fabs could supply the world's needs, if the money is there to support them.  But without growth and cheap energy (e.g. fossil fuels), money will be in short supply, so things will get simplified.

But how simple?  A basic transistor or two enables radio receivers, which are quite valuable
for news, weather, farm prices, entertainment, etc.  A few more can make a simple transceiver (think CB radio), adding two way communication, meaning aid can be summoned faster, trades can be discussed faster/easier - especially in light of the reduced mobility likely in a fossil fuel free world.

A few power transistors and some simple ICs make an efficient, reliable, life conserving charging/management system for storage batteries, which extends the utility of small wind turbines, PV systems, solar thermal electric, wood burning Stirling engines, and the like.

A few more and bigger power semiconductors, and one can build inverters, and run AC motors, which means longer life/higher reliability due to the replacement of commutators and the attendant brush wear with slip rings, or with permanent magnet motors, not even any slip rings (by electrically commutating the stationary windings on the outside of the motor).  On a per-motor basis, like an electric buggy/bike/train/..., one can do efficient speed control and regenerative braking.

Simple, reliable motors mean more effective things like refrigeration, circulation pumps for solar hot water and ground source heat pumps, water pumps for irrigation, fans for ventilation, and I'm sure many more applications.  In conjunction with some simple electronics, motors can be used to track the sun for solar collectors, which means higher temperatures for solar thermal systems and more output per day for all solar collectors.

Another source of motivation is that semiconductors do have "wear-out" mechanisms, depending on how they're used.  Contaminants eventually leak into packages, causing corrosion and shifts in electronic properties.  Electromigration due to the flow of electrons displacing the atoms in conductors causes open circuits, and electrostatic displacement of atoms can cause short circuits.  Diffusion of dopants and contaminating impurities within the solid semiconductor crystal will in time alter the device characteristics, resulting in malfunction.  While commercial semiconductors can be made to work reliably for some 10s of years, say 20 or 30, when we get into the 100 year or longer timeframe, failures will be quite high. [4]  Also note that the "passive" components used in electronics systems, like resistors and capacitors, also have failure mechanisms.  Thus unless we replace broken devices, it won't be long until we won't have anymore.

The Evolution Of The Semiconductor Industry

The solid state transistor was invented in 1947 [1], it was a very crude device, requiring no special environment to make.  Since it was fragile, other kinds of transistors were developed through the 1940's and 1950's.  But these were still made in ordinary environments without that much work at extraordinary cleanliness. Transistors were initially made individually, mostly by hand.  Around 1958/59, with the application of photolithography techniques from the printing industry, together with diffusion doping, multiple transistors started being made on a single wafer, which was then cut into separate "die", each with one transistor on it.  Shortly thereafter, multiple transistors were made and hooked together on a single die and the integrated circuit (IC) was born. [2]

The very first ICs had fairly crude dimensions, lines on the order of 100th of an inch (.25 mm), easily achievable with photolithography, and not too susceptible to things like dust particles, though hairs and lint could be problematic.  So the first "clean rooms" started to be in use for electronics.  Clean room techniques were developed a bit earlier, especially during World War II to deal with miniature ball bearings and other tiny components. [3]

Air conditioning is a must for clean rooms, both to circulate air through fine filters and to maintain a reasonable temperature, since any heat inside the room must be removed artificially because letting in fresh air will just dirty the place.  And in most clean room applications, air conditioning is required to maintain a constant temperature due to the small tolerances of the items being made, which would be overwhelmed by thermal expansion/contraction.  Without temperature control, items made early one day would not fit items made in the hot afternoon.  Even a single wafer being processed must match a series of masks, which would be different relative sizes at different temperatures.

Cleanrooms come in different "classes" according to the need.  The latest microprocessors require class 1 (ISO 3) or better cleanrooms. [5]  Class 1 means no more than 1 particle measuring bigger than .5 um in each cubic foot of air (35 per cubic meter).  This is an enormous undertaking.  The cleanroom must be positively pressurized, to keep any dirt out.  The air blows down from the ceiling and out through the floor to sweep particles away.  Very fine filters, with large pressure losses, filter all the air many times an hour.  Redundant filter setups are required to allow changing filters without losing cleanliness.  Everything that goes into the cleanroom must be thoroughly cleaned.  We've probably all seen some picture of the "bunny suits" warn by technicians in modern fabs.  These must be made of synthetic fabrics that won't shed any lint.  Along with hoods and booties, they too much be cleaned before use and maintained.  Some fabs require fully contained suits, filtering the wearer's breath.  To help keep contaminants off wafers, they are now transported from machine to machine inside Front Opening Unified Pods (FOUPs).  All these things are made of plastics, meaning petroleum.  And many are disposable, especially the cleaning wipes and sooner or later the suits.

In contrast, PV fabs barely have cleanrooms - class 100,000 (ISO 8), the lowest cleanroom class - to keep metallic dust down that would cut a percent or two off the efficiency of the cells.  Basically some better than normal air filters and paying attention to ductwork for a regular air-conditioning system.

A property of ICs is that by shrinking the transistors and interconnections, one can trade off putting more on a single chip (increased function) or making more chips per wafer (decreased cost per function).  Since integrated circuits have to be designed into a system, IC manufacturers want to be at the leading edge of features, so they can have more "design wins" than their competitors.  While this is costly up front, as a generation of ICs is produced, experience is gained, yield goes up, cost goes down, and profits go up, especially at high sales volumes.  So there is great pressure to shrink feature size.  In 1971, Intel came out with their first microprocessor, the 4004.  It had 10 um features and 2,300 transistors.  A sheet of 20 pound typing paper is about 100 um thick.  Blue light, used in the photolithographic process, has a wavelength a little less than .5 um, so 10 um features are easy to see under visible light, and easy to use conventional photolithographic "masks" (like a photographic negative) to expose the patterns needed.  By 2000, the Pentium 4 had 0.18 um features and 4.2 million transistors, and in 2012 the 3rd generation Core processor has 0.022 um (22 nm) features and 1.4 billion transistors.  The diameter of a single atom is around 0.1 nm, so chip features are now just a few 100 atoms big. [6]

Photolithography of semiconductors started simple.  A wafer was chucked onto a spin-able plate, then drops of a solution of photoresist were dropped onto the center.  As the wafer is spun, this will spread out into a uniform thin layer and harden as the solvent evaporates.  After letting the resist dry, a full wafer mask would be aligned and set face down on the wafer and exposed to blue or UV light.  Then the resist is chemically developed, leaving some on the wafer and other areas where it was removed.  Now the wafer could be exposed to etchants that  remove unwanted areas, or to additive processes to create structures only where wanted.  Soon masks became so complex, and tolerances so fine, that a mask for the whole wafer was no longer economical or gave good yield if actually pressed onto the wafer.  So "step and repeat" machines were developed that exposed each chip's area individually by projecting light through the mask.  This also relaxed mask constraints so the masks could be made larger than the resultant feature size. [7]  These days features are much smaller than visible light, so ultraviolet light from excimer lasers is used, as well as tricks like phase-contrast masks to get sub-wavelength resolution.  These steppers are very complex and very expensive, $40 million each, and a large fab will have a dozen or so. [9]  Even the excimer lasers aren't fine enough for smaller features, so extreme ultraviolet lithography is being developed. [8]

While the costs per wafer are going up, the costs per transistor have been going down way faster, but it takes a large amount of capital to build and equip these plants.  PV wafer processing cost about $1, while a 300 mm wafer might cost $3,000 - $10,000. [10]  Even accounting the difference in wafer area, $3 to $3,000 means there's a lot more complexity on a semiconductor wafer.  A simple recipe for an IC has more than 20 steps, here's one from a university [11].  It lists most of the supplies, chemicals and equipment needed, including the 4 masks.  The early masks were "taped out" by hand [13] and photographically reduced with ordinary high resolution film.  Now, a complex process for state of the art microprocessors will have more than 300 steps, and a set of 30 or so masks. [12]  The features on masks are now so small that electron beams are used to "write" them.  The finer features/small light wavelengths require new photoresists, which require new solvents to apply them and new solvents to dissolve them.  Other layers may be required/desired below or above the actual photoresist layer. [15]  All these chemicals keep getting more and more complex.

Doping got more complex.  It started out as simple diffusion from gaseous, solid, or spun on precursors.  These days, ion implant machines, basically particle accelerators, blast dopant ions into the wafer.

As feature sizes decreased, simple wet etching needed to be replaced with dry plasma etching processes to avoid problems with microscopic gas bubbles blocking etchant action and to achieve the higher feature vertical aspect ratios desired.

Metallization to interconnect transistors has gotten vastly more complex.  The early metallization was just aluminum, evaporated or sputtered over patterned photoresist.  When the photoresist was washed off in solvent, aluminum was left behind on the bare silicon or oxide surface.  But when chip features got very small, two issues emerge.  Since aluminum is a light element, when conductors get small, it is subject to electromigration of the aluminum atoms from the pressure of the flowing electrons, resulting in open circuits.  The solution was to replace aluminum with copper, but that requires more steps, since putting copper directly on silicon will "poison" the devices, and copper can't be dry etched and is hard to evaporate or sputter, so it must be electroplated.  Since it can't be dry etched and wet etching is no longer fine enough, it must be over-deposited in trenches in the insulator around it, then chemical-mechanical planarization (CMP) to pattern, which requires more steps, and more chemicals/materials.  [14]


The current microprocessor/memory technology has become very complex, expensive and depends on complex supply chains.  While certainly important to the military industrial complex, the large market for consumer goods might shrink considerably in a declining economy.  And since the military depends to a large degree on Commercial Off The Shelf (COTS) items now, we all may be blindsided by a collapse of the IC industry or large parts of it.  Also, with a more fragile economy, recovery from things like the Fukushima earthquake may be more difficult/impossible - it had a significant effect on the IC industry. [16]

Processes for Simple Semiconductors

I will lay out some processing steps for simple semiconductors, using something like 10 - 20 um features ("rules"), which should be sufficient to make transistors and simple ICs like voltage regulators, op amps, and the like.

The first thing different from the silicon PV processing described in part 2 are more exacting cleaning requirements.  A simple dip in hydrofluoric acid (HF) will not do, since transistors are much more sensitive to contamination.  The standard IC clean starts with acetone, then isopropyl alcohol (IPA) , then a deionized/distilled (DI) water rinse.  Sources of these have already been covered, but the purity levels would need to be stepped up a notch, including freedom from particulates.  Natural fiber filters would work on these, but other things would require plastics.  A brief dip in dilute HF is to take off the native oxide that will form on silicon exposed to air. The next step depends on a hydrogen peroxide/ammonium hydroxide mixture.  Hydrogen peroxide has many pathways to make, and hopefully there is a large amount being made for paper bleaching, if not, the barium peroxide + acid method should be easy at small scale and the byproduct recyclable.  Ammonium hydroxide is just ammonia gas bubbled through water.  After that, comes another dip in HF to strip the oxide formed.  The next solution is hydrogen peroxide/sulfuric acid.  A fair amount of DI water rinsing is required, so some attention to water treatment/reuse will be required.  Also, personnel protection is an issue, so a supply of rubber gloves and aprons is needed.

Among the attributes of silicon that make it the semiconductor of choice is its stable, insoluble oxide.  So the first real processing step is to grow an oxide.  Oxides serve as masks for doping/implantation, insulating/passivation layers, and gate oxides for simple MOS transistors.  One needs a furnace that goes to 1100 deg. C (2000 F) or more, with a sealed chamber - typically a fused quartz tube.  Fused quartz rods are also used to build the "boats" that hold the wafers in the furnace.  A supply of oxygen, hydrogen and nitrogen gas is needed.  The nitrogen keeps a flow of gas always going out of the furnace, to try and keep contaminates out.  The oxygen and hydrogen are used in oxidation, the oxygen by itself makes oxide with different qualities than oxygen and hydrogen together, which make high temperature steam - a "wet" oxidation.  Often these are mixed in the same run.  Depending on the thickness of oxide layer desired, oxidation may take half an hour to several hours.

Once an oxide for masking is grown, photoresist must be laid down.  The wafer must be completely dry, so a minute on a warm hotplate is often used if it has been sitting around adsorbing water from the air.  Often a surfactant/bonding agent is used between the silicon and photoresist, often hexamethyldisilazane.  I would hope to skip this step for simple things, it's a kinda nasty chemical, but if we prepare silicones for PV encapsulation, it can easily be made.  A wafer is chucked on a spinner.  These have a vacuum chuck to hold the wafer down, and a motor and often a timer to quickly spin a wafer to several thousand rpm.  Then we drop on some liquid photoresist and spin it off.

There are two kinds of photoresists, positive tone and negative tone.  The positive resists dissolve where the light hit them, and the negative resists dissolve where the light did NOT hit them.  There are many kinds out there, so I'm reasonably sure something could be found.  There are all kinds of tradeoffs in choosing a resist.  Fortunately not much of it is needed.

Masked exposure is next.  Contact masks, where the mask is actually touching the resist on the wafer, are simplest, but have obvious wear and contamination problems.  Masks are made from glass or fuzed silica with a corrosion resistant metal on them, often chromium.  They are patterned photolithographically themselves, usually with a reduction lens from a larger pattern (at least for something like 10-20 um rules).  The light could be a mercury vapor light, or with the right resist, the sun.  Simple steppers could be used if the tradeoff from making many contact masks vs. the stepper works on in favor of it.

Resist development follows, where the dissolvable parts of the resist are washed away.  Some developers are as simple as a mild solution of sodium hydroxide, others are very complex.

After development, one typically needs to hardbake the resist to help it stand up to the following processing steps.  This is often done on a hotplate or small oven.

An alternative method of patterning might be nano-imprint lithography (NIL), though not at the nanometer scale.  Instead of photoresist, a warm, flowable polymer (or a UV curable liquid) is squeezed between a mold and the wafer surface.  When cooled (or cured), the mold is removed.  The polymer could be something as simple as beeswax.

Now one does an oxide etch, using HF and often an ammonia salt of HF, ammonium fluoride, easily made from HF and ammonium hydroxide.  This will etch holes in the oxide down to the silicon surface wherever the resist is not covering the oxide.

Often, the photoresist is removed before the next step, particularly if a high temperature step like diffusion follows.  This can be done with solvents, like acetone.  Now we have a waste, but we can distill that to recycle the acetone.  The other way is "ashing" - using an oxygen plasma to burn the photoresist. [17]  Typical RF requirements are 50 W at 30 kHz, well within simple vacuum tube or transistor capabilities.

Dopant deposition is the next step.  There are several alternatives.

Solid source doping uses wafers made of dopant precursors or have dopant precursors deposited on them.  They are place face to face with the masked wafer and heated so that dopant atoms diffuse to the masked wafer. The advantages are the usually non-toxic nature of the source.  It requires a hot furnace step, then cooling before the drive in step, so it is slow.  It's also not so controllable, due to the depletion of the dopant source wafers.

Gas source doping uses a gas (or volatilized liquid) in a hot furnace.  The dopant decomposes on the hot wafers.  It requires fairly sophisticated flow controllers to be repeatable.  It allows twice the wafers in a furnace as solid source doping, and one can often just shut off the dopant gas and then raise the furnace to 1100 - 1200 deg. C to do the drive in step.

Spin on sources are just liquids that are spun onto the wafer like photoresist, then baked in a hot furnace.  Typically it must cool and the "spin on glass" removed with HF, then back into a diffusion furnace, adding extra steps.

Drive in occurs when the wafers are heated to 1100 or 1200 deg. C, and the dopant atoms rapidly diffuse inward.  There are usually several dopings, so the "thermal budget" must be planned out in advance to avoid over diffusion of the first doping.  Often the masking oxide is removed just before this step, and another fresh oxide layer grown at the same time, in preparation for the next step.

An alternative is ion implantation.  If you'd ask me a few years ago about low tech semiconductor fabrication and ion implantation, I'd have laughed.  But there are some low cost implanters that have come out for PV, so I no longer think it that wild.  The PV style implanters don't mass analyze the ions, nor do they raster the beam, so they'd only work for unsophisticated electronics on small wafers - just like I'm thinking of here.  An advantage is that if the resist layer is thick enough, no oxide is needed to mask the implant.  One can use a rapid thermal process to anneal and activate the implant instead of an hour(s) long diffusion.  It would save creating and removing the oxide, and be better controlled than other doping methods.  Basically an ion implanter is a high current particle accelerator, like the Cockcroft-Walton invented in 1932, although the 50 - 100 keV needed is well within the range of a transformer and rectifier.

Once the drive-in step is done, if one has used an oxide mask and it is still there, one must remove it in HF, unless one is keeping it as the final oxide protective layer.

There are typically 2 or 3 diffusions/implants depending on what one is doing.  Simple transistors only take 2.  ICs will take sometimes take 3, as will thyristors and similar devices.

If doing metal oxide semiconductors (MOS), a gate insulator is needed.  For simple devices just grown an oxide layer, then do photoresist, a photomask, develop, hard bake, etch the oxide in HF, the remove the resist.  Gate oxides need to be of high quality, so they are typically grown "dry" with just oxygen, but that is much slower than "wet" oxidation.  Many process flows will leave the last "wet" oxide on the surface as a protective layer, etch down to bare silicon for the gate, then grow the gate oxide in those holes.

Next we need to etch off any oxide where we will put down metal.  Typical step as usual: spin on photoresist, expose photomask, develop, hard bake, etch oxide in HF, remove the resist.

Metallization is most simply done with evaporated aluminum.  Aluminum is compatible with silicon, and for crude devices like we're making with 10 or 20 um rules it's reliable.  It also works well with a process called "lift-off", where a layer of photoresist is laid down and patterned, leaving resist where we do NOT want the aluminum to end up.  Then we evaporate a layer of aluminum, then dissolve the resist and any aluminum on top of resist will come off too.  We could do a masked etch of the metal instead of lift-off, but now we need a metal etchant in our toolbox, which already has acetone in it, a good solvent for lift-off.

The evaporation machine is a sealed vessel, pumped down to a decent vacuum.  The simplest ones are just a tungsten filament on which little aluminum wires will be hung.  After a good vacuum is achieved, the filament is energized a bit and the aluminum wires melt and form balls by capillary action.  Then the filament is heated more, and the aluminum evaporates at 2519 deg. C., far below the melting point of tungsten at 3422 C.  Fancier evaporators use a ceramic crucible or electron beam to melt/vaporize aluminum. We'll put down about a um thick of aluminum of small signal devices, more on big power devices.

If we had a sputtering machine, we could use that instead of evaporation.

Lift-off requires using a solvent, like acetone, to reach in through the stepped edge of the aluminum layer.  Ashing will not work at this step.

The final step when doing aluminum contacts is to anneal the contacts, at 475 deg. C. in an inert or slightly reducing atmosphere.  This improves the electrical conductivity of the contact.

Now we test each device.  A prober will have a set of needles that can be lowered to contact each of the device input/output pads, and an electronic circuit to test the function.  This could be very simple - an function generator for input, and oscilloscope for output, and a picture of what's good or not.  The operator will mark bad die so we don't waste resources packaging them.  The prober will then be retracted and positioned over the next die.

The wafer is then diced.  These days most of sliced up with a dicing saw, which uses small diamond coated wheels.  With the right orientation of wafer, we can scribe with a diamond or tungsten carbide tipped scribe, and break the wafer into lines, and each line into die.

Then the die will be packaged.  These days most die are packaged in plastic, but I'm doubtful about that in a future petroleum short world.  The technology in the resins is actually pretty fancy.  Instead, we use plain old-fashioned metal or ceramic packages.  For small transistors, like the TO-18 for small ones, and TO-3 for large ones.  ICs with up to 8 or 10 leads can use the similar TO-99 package.  These are made of metal, with leads going into the base through hermetic metal-glass seals.  The die is attached to the base of the package, sometimes on an insulator, with a low temp eutectic solder or glass frit.  Then wires are run from the die to the metal leads.   The simplest way involves tiny gold wires.  These are obviously expensive, but easy since the gold is so soft.  A thin gold wire is fed through a tiny metal pipe like a hypodermic needle (with a flat, smoothed end).   A bit of wire is fed out through the tube and balled.  Formerly a tiny oxy-hydrogen flame was used, but these days an electrical discharge melts the tip of the gold wire, forming a ball.  Then the ball is pressed onto the IC pad and pressure welds to it.  The wire is draw over to the package lead, and smashed into that, bonding and pulling the wire apart. [18]  Copper and aluminum are also used for IC wire bonding, but have to be ultrasonically welded. [19]  After die bonding, it is common to test again, to make sure the attachments are good.  Then the cap is put on and sealed with a low temperature solder.

Discussion

If you've made it this far, congratulations.  The manufacture of even simple transistors is non-trivial.  But nothing was extremely hard, or depended on sophisticated tools/controllers/exotic materials/complex infrastructure like modern microprocessors do.  A simple class 100,000 or 10,000 cleanroom should work well enough for transistors and simple ICs that would serve as simple radio amps, battery charge controllers and the like.

No sane military commander would go off to battle without two way radio if he could help it, and although tubes are much more low tech, their power requirements are much greater and that is problematic for mobile applications.  Radar and night vision equipment is also of great benefit for military use.

Having a good battery charge controller would greatly extend the life of storage batteries, and with small wind and/or solar being the only two near-universally applicable sustainable electricity generation means, there would be great economic benefit in that.

LEDs cannot be made from silicon, but the development of skills and infrastructure would greatly facilitate the practice of that technology, which uses similar equipment.  They would reduce power requirements for lighting by 80% over incandescent bulbs, and save even more money/effort/resources due to their long lifetime.

No promises can be made, but simple semiconductors don't look impossible, even in a somewhat resource constrained world.

============== References =============
[1] http://en.wikipedia.org/wiki/History_of_the_transistor

[2] http://en.wikipedia.org/wiki/Invention_of_the_integrated_circuit#First_semiconductor_ICs

[3] http://www.academia.edu/1059377/Controlling_Contamination_The_Origins_of_Clean_Room_Technology

[4] http://www.aosmd.com/media/reliability-handbook.pdf

[5] http://www.intel.com/content/www/us/en/history/museum-transistors-to-transformations-brochure.html

[6] http://en.wikipedia.org/wiki/Atomic_radius

[7] http://en.wikipedia.org/wiki/Stepper

[8] http://en.wikipedia.org/wiki/Extreme_ultraviolet_lithography

[9] http://www.bizjournals.com/albuquerque/stories/2009/07/20/daily73.html?page=all
And notice the power consumption for this fab - 55 MW going to 65 MW.

[10] http://smithsonianchips.si.edu/ice/cd/CEICM/SECTION2.pdf

[11] http://fabweb.ece.uiuc.edu/lab/

[12] http://en.wikipedia.org/wiki/Photomask

[13] http://www.computerhistory.org/semiconductor/timeline/1955-Photolithography.html
Note 3rd and 4th pictures.

[14] http://en.wikipedia.org/wiki/Copper_interconnect

[15] courses.ee.psu.edu/ruzyllo/ee518/EE518_Adv.PR.Tech.S06.ppt

[16] http://www.digitimes.com/news/a20110315VL202.html
http://www.digitimes.com/news/a20110322VL200.html

[17] http://en.wikipedia.org/wiki/Plasma_cleaning

[18] http://en.wikipedia.org/wiki/Ball_bonding

[19] http://www.smallprecisiontools.com/products-and-solutions/chip-bonding-tools/wedge-bonding-tools/technical-overview/basic-ultrasonic-wedge-bonding-process

Wednesday, October 30, 2013

Krampus Wish List - Photovoltaics



Krampus Gift List - Photovoltaics

This is part 2 of a (hopefully) 3 part series in response to
http://thearchdruidreport.blogspot.com/2013/01/a-wish-list-for-krampus.html

I cover photovoltaics (PV) in this article, in respect to a "post-industrial" world where fossil fuels are depleted and humans must use sustainable sources of energy.  PV - the direct conversion of sunlight to electricity - offers a sustainable supply of electricity, provided one can produce the PV cells and connect them into useful modules.

Terminology

  • volt - the unit of electrical potential or "push".  A small single battery is usually about 1.5 volts, barely enough to make a tiny spark.
  • amp - short for ampere (A), the unit of electrical current, or quantity of flowing electrons.
  • watt - the usual unit of power, that is the rate of energy flow.  One definition is the power of 1 amp flowing through an electrical potential difference of 1 volt.  Roughly, 100 watts is the power given off as heat by a 100 watt incandescent lightbulb, which at 120 volts feed would draw just under 1 amp of current.  My big toaster is 900 watts (W), or 7.5 A at 120 v.
  • watt hour (Wh) - unit of energy, the product of rate of flow times time of flow.
  • AC. - alternating current, electrical current that reverses polarity, most at 60 (in the U.S.) or 50 (in Europe) times per second.   The standard wall outlet, lighting current.  The alternation allows the voltage to be raised or lowered by inexpensive means of transformers.
  • DC - direct current, electrical current that only flows in one direction, like batteries or PV modules.  Most electronics devices use D.C.
  • watts peakWp - the output of a solar cell/panel under the standard test conditions of light intensity, spectrum and temperature.  The light intensity is defined so that at noon in a temperate latitude, the power of the sunlight is 1000 watts/square meter exactly facing the sun.
  • cell - the smallest unit of PV, one photodiode, it produces less than one volt, current depends on area and efficiency.
  • strings - used for either cells or modules electrically joined together.
  • module - often "panel", one or more cell strings encapsulated, often with a frame.  The smallest installable unit of PV.
  • array - one or more modules, in one or more module strings, hooked together to provide DC current when the sun is shining.  May be hooked directly to batteries or a DC load, but most often to an inverter and/or battery charger.
  • inverter - electronic device to convert DC current into AC current.

I will use k for kilo- (thousand), M for Mega- (million), and G for Giga- (billion 10^9)

State of PV in the World

Many people are surprised by how big the PV industry's impact is these days.  In 2012 the U.S. installed 11% of the world's newly installed PV, 3,313 MWp, so worldwide that was roughly 30 GWp installed that year. [24]   3.3 GW is about 3 large conventional power plants, though the capacity factor of PV is only about 20%, due to the variability of sunlight, so the output will be less.  But the point should be noted, PV is starting to make real impact in a big way.  At the first of this October, Germany hit 59% renewables on a cool, sunny, windy day, with PV providing a substantial portion, peaking at 20.5 GW. [25]  That's a big amount of power in anybody's book.  In Feb. 2013, it's estimated that cumulative worldwide PV passed the 100 GWp mark. [27]

100 Wp is a small "standard" panel for crystalline silicon, the smallest one in the picture of these Kyocera modules is 140Wp.  It is 59.1 inches (1,501 mm) long x 26.3 inches (668 mm) wide, at max power output gives 7.9 amps at 17.7 volts.  It's module area is 1.003 square meter, so 140 Wp means it's 14% efficient at the module level.  The biggest one to the left is 320 Wp.

A U.S. house uses about 30 kWh per day rough average, though that can be reduced considerably.  So with about 6 hours of full sun/day, a 5 kW (effective) array would supply enough electricity for daily use.  I say "effective" due to things like orientation of module, shading, etc.  And either net metering would need to be used to trade electricity with the grid, or a bunch of expensive batteries used to store energy for nighttime.  If one reduced usage by half, the PV system - modules, mounting, inverter, wiring - would accordingly be half.   Using the 14% efficient modules above, we'd need 35.7 square meters (385 square feet) for the 5 kW system.  Given the average new U.S. house is over 2000 square feet, we'd have plenty of room to put PV modules on the roof, alongside a solar hot water array.

Small (< 10 kW) PV systems in the U.S. are now costing $5.30 per Wp on average, a significant decline from the $12/Wp in 1998.  In Germany, small systems are installed for $2.60/Wp. [26]

So, PV is here, is able to provide adequate amounts of electricity for pretty much any task. Intermittency is solvable with storage, though at increased cost.  Costs are coming down, and conventional utility rates are rising.  Hawaii is already at retail grid parity, as are certain commercial customers in California subject to peak rates.  But there's a long way to go to Terawatt levels of production that would replace most conventional generation.  Can we get there in time with PV and other renewables?

Possible Uses in a "post-industrial" World

I start with some possible uses of PV to give one an idea of just how useful even small amounts of electricity can be.  First is lighting.  Imagine hunting or gathering, or farming, or traveling for trade, or seeking aid (perhaps for a medical/animal emergency) or going to someone's aid, and it turns night.  Imagine the wind is blowing.  Imagine trying to stumble through forest or field, or even along the rough roads of the future, without a light, or a crude torch that keeps going out in the wind.  Did you remember to carry a fragile oil lamp even though you were sure you would be home by dark?  Would an oil lantern be as convenient as a headlamp when farming after dark?  Could you make out that animal in the dark with a flame based lantern as opposed to a flashlight beam?  If you were attending to a injured person after dark, would a candle let you see well enough?

In our modern world, where a light is a switch away, and even most streets and many highways are well lit, it's hard to really comprehend what life without electric lights is about.
The third world has plenty of experience, relying on kerosene lamps with their pollution, fire danger and expense; and is welcoming PV powered lighting. [1]  With LEDs, one can get 8 hours of a small light with just a few "watts peak" (Wp - the output of a solar cell/panel under the standard test conditions of light intensity, spectrum and temperature) of PV.  Many of you may have seen the solar garden lights - something a bit brighter would be plenty useful.

Kerosene lanterns have been in use since the 1850s, and were the first demand source for crude oil.  Besides the health impact, [3], they use more than 1 million barrels/day of kerosene, more than 1% of world oil consumption, and that will not be possible much longer as peak oil hits.

A side note of sorts, using efficient LEDs, a group has developed a gravity powered light. [2]  This has the advantage over PV of needing no batteries, which will need replacing over the years.  It's only practical for stationary applications, and doesn't seem to charge mobile phones..., but the idea has legs.  As someone who has tried windup/crank/squeeze flashlights, I think the gravity light is a great advance.

The next use of electricity to consider is radio.  This can be a simple, receive-only radio that has been in use since the 1920s for news and entertainment.  If people are dispersed to the countryside as farmers, they may not be able to make it to town to get a newspaper, even if there were newspapers (and they could afford them).  The weather, crop prices, and local developments that affect them (say a road closure due to bridge washout, some traders arriving in town, an epidemic of some kind, etc.) can be very valuable in avoiding excess losses and wasted time.

These simple radios could be powered (as many were) by simple wind turbines charging batteries, but there are some parts of the world that have no wind for days at a time, whereas the sun comes up every day.  These kinds of radios need only a few tubes or transistors to function very well, and can be quite cheap.  Tube radios would need a couple of watts, but transistor radios will work on even less power. The other end is the transmitter station, which also needs electricity, though more of it.  A few hundred watts on up for a tube station,
50 to 100 watts or more for transistors, for a simple AM station with a 10 or 20 mile reach.

Next up in complexity would be two way radios, like CB radios and walkie-talkies.  Tube designs would have limited applications and require fairly large PV arrays to power them, but transistor walkie-talkies and CB radios could be charged on 10-30 Wp of PV.

Refrigeration would come next on my doomstead wish-list.  While canning, root cellars, drying and the like work for a great many foods, refrigeration and freezing work for pretty much any food item, including fresh made batches of things like soups, beans, cooked grains, cheese, etc.  Much time can be saved by cooking a large batch of these sorts of foods, and simple re-heating (or not) at mealtime.  Note that climate change means most areas will be getting warmer, and so food will spoil faster.   Another use is for medical purposes, keeping medicines (including herbal preparations) cool, and providing cool for fevered patients.  An interesting book that deals in this is The Frozen Water Trade, which details the distribution of New England ice by sailing ships as far as Calcutta India in the mid 1800s, and the life saving use of ice for tropical fevers.  Some of the solar refrigerators will work off PV arrays as small as 100 Wp, given sufficient insolation. [4]

I might mention that solar thermal refrigeration is also a possibility, using heat from the sun to drive an absorption refrigeration cycle, as opposed to a vapor compression cycle. [5]

Closely following refrigeration might be water pumping, both from wells or from surface water.  While it may be romantic to think of getting a bucket or two of water from a manual well pump, how to get it to your roof for your solar powered hot water heater?  Or pump hundreds of gallons to irrigate some crops?

Once water is pumped, it might be nice to sanitize it, since we're running out of antibiotics that still work to fight infections.  A small air pump and ozone generator will kill most things, and/or a UV light will stifle further growth of living organisms including viruses.  It would be nice to have a disinfectant residue like hydrogen peroxide or chlorine.  Both of these can be generated via electrochemical processes, at fairly small scale, but the peroxide generators require fairly rare metal catalysts and somewhat exotic membranes, and the chlorine/bleach generators are rather fussy, so maybe this isn't a household thing.  But if you knew about such, and had lost a child to dysentery, you might want to get such a unit and the means to power it.

Fans are another PV power-able item it might be handy to have, especially as climate weirding increases.  Small fans can draw just a few 10s of watts.  Full out air condition might be pushing it, but it certainly is possible, especially with efficient ground source heat pumps, though the PV array required gets into the kW range.

Once one gets into the kW range, then electric cookery is possible, as are electric motor based tools.  Electric cooking can be done with slow cookers (though solar typically works fine if you've got the location/sunlight), resistance heaters (toaster, conventional electric stoves), induction cooktops (will need transistors as well), and even microwaves (the cavity magnetron is a surprisingly simple device).

And once one gets in the 10 kW range for a PV array, things like induction forges become possible, for a low carbon, low impact metal industry.  Once one gets to the 100kW or MW range of PV arrays, then one can do things like ... make PV panels.

Approaches to PV

I will mainly discuss crystalline silicon based PV, but first want to give a bit of background on other approaches, most of which are less likely to be around in the future.

One hears (in certain circles) a great deal about organic PV, and how one will be able to just paint your house with some special organic coating and voila!  Well, that is wishful thinking.  For one thing, the photovoltaic effect will only give a voltage up to the "band gap" of the material, and with visible light, that maxes out about 1 volt, typically .6 volt or less.  The looses due to resistance would eat up all the power.  Secondly, all organic PV (dyes or "small molecules") are sensitive to oxygen and water, especially in sunlight.  So for any lasting use, they must be carefully sealed, but even then device lifetimes are on the order of 1000 hours.  Thirdly, especially for dye cells, but also for "small molecule" organic semiconductors, the chemistry is considerably more complex that what I've outline for silicon.

Amorphous silicon (a-Si) is derived straight from silane gas, and its adherents tout the efficient use of that chemical in laying down a thin film (silane can be made from tri-chloro-silane too).  But, a lot is wasted in the deposition process, and the resultant efficiency is low (6%), the long term stability of the a-Si is low and it too has to be carefully sealed.  One finds a lot of it in consumer items, like "solar powered" calculators, but most of them don't spend much time in the sun.  At one time the a-Si guys had a lot of hope, but in the last few years most of the companies producing this have gone bankrupt - unable to compete with plain old boring crystalline wafer silicon.

Copper Indium Gallium Selenide (CIGS) is another thin-film technology, infamous as the choice of Solyndra, the bankrupt PV company.  Other CIGS companies have survived for now, but the consensus in the industry is that outside a few niche markets (flexible, lightweight PV) it just can't compete for big power applications.   CIGS is a very moisture sensitive material, and indium is fairly rare.  While champion efficiencies are about the 20% range, real world modules are closer to 12%.  If one did stumble upon an indium mine, and had access to good polymers for encapsulation, then CIGS might be worth a look.  The inputs are soda-lime glass, molybdenum, copper, indium, gallium, selenium, cadmium, sulfur, zinc oxide or indium tin oxide, some kind of anti-reflective coating like MgF2, nickel or a similar metal, and aluminum for front contact.  Then a good polymer encapsulant and another sheet of glass.  One does a lot of sputtering, so argon gas and vacuum pumps will be needed.  Copper Zinc Tin Sulfide (CZTS) is a similar material, but at a lessor state of development.  CZTS does have the advantage of not using the rare metal indium.

So that leaves cadmium telluride (CdTe) as the sole competitor to wafered silicon PV.   Even this is questionable as the cost of silicon keeps coming down, but for now it has the (barely) cheapest module costs.  CdTe uses both a toxic (cadmium) and a rare (tellurium) element, so there are questions, but frankly somewhat overblown on the toxicity issue.  The tellurium supply issue is open, but worrisome.  Efficiencies are around 12%, comparable to the lowest crystalline silicon on the market.  Both CdTe and CIGS have several manufacturing options, which could be simpler at small scale (assuming raw material availability).   The low efficiency means more glass area is required, and the high moisture sensitivity means glass is a must on the backside as well.  Also, thin films are made in monolithic modules, which means one bad cell requires junking the whole module.  This would also apply in the field.  A glass-glass monolithic module is essentially impossible to fix, whereas even a glass-glass module made of wafers could be drilled from the back to isolate bad cells and jumper around them.  If times are really tough, fixing what one has may be the only option to doing without. 

Making a Silicon PV Cell from a Wafer

I describe a typical process of the standard silicon solar cell.  It is made on a p-type wafer that is typically doped with boron (or could be doped with gallium).  This won't make the highest efficiency, but is the easiest process, accounting for the vast majority (90%) of solar panels in the world today.

First I want to say that PV is not made in "chip fabs" that make integrated circuits for computers, phones and the like.  Aside from university research projects and pricey multi-junction compound semiconductor cells for outer space, nobody in the PV world uses photolithography - all patterning is done with screen printing (the vast majority), ink jets, lasers and the like.  The cleanliness requirements for modern IC processing are literally orders of magnitude more strict than for making PV cells.  One could make silicon cells in one's garage starting from raw wafers if one had a ventilation hood, a hotplate, some suitable processing containers, and an oven that reached 1000 deg. C. (1,800 F).

Step 1 is cleaning/saw damage etch.  Depending on how greasy/dirty the incoming wafers are, one might have to pre-clean with solvents.  Ethanol, acetone, pretty much anything will do.  The solvents could be reclaimed with a solar still.  The main goal though is to remove the saw damage layer, which is about 10 um of cracked silicon contaminated with metals from the sawing process.  This can be combined in a single step using a hot (nearly boiling) and concentrated (around 10 molar) solution of sodium hydroxide.  The sodium hydroxide can be produced as described in part 1.  Sodium hydroxide will slowly etch glass, so it would be greatly preferable to have some plastic container to do this in.  Many plastics would work, polyethylene, polypropylene, ABS, PVC are common ones that one could find as salvage.  Otherwise alumina and a few other ceramics could be used.  I do note that there is a process to make polyethylene out of sugar cane ethanol. [6]

We need some distilled or deionized (DI) water to mix things up and rinse them.  A solar still would work for the distillation.  Put on your goggles, face shield, rubber gloves and apron, then mix the sodium hydroxide (lye) into water.  If your hotplate doesn't do good temperature regulation, use a double-boiler arrangement.  A solar hot water installation could at least pre-heat the heating loop.  Heat a container of sodium hydroxide, load some clean wafers into a basket-like carrier, and dip them in the hot sodium hydroxide.  Set a timer - I recall on the order of 10-20 minutes for this step.  When done, rinse in (DI) water.  The waste contains sodium silicate (water glass), unused lye, silica, and tiny amounts of trace impurities.  You could let it evaporate a bit and reclaim the lye by recystallization, but eventually the silica will mostly precipitate.  You can feed that to the silicon process after washing it, and the impure lye - I dunno, make some soap.

Step 2 is to texture the surface of the wafer for better light trapping, which will increase the efficiency (and decrease use of encapsulation/mounting materials).  For mono-crystalline boules, a lower temperature, more dilute solution of sodium hydroxide is used, along with a surfactant, often isopropyl alcohol (IPA).  Other surfactants could be used, or an old process of fermentation used to produce acetone [7], which can then be hydrogenated into isopropyl alcohol (hydrogen from electrolysis of water, Raney nickel or similar catalyst).  This solution preferentially etches based on the different crystal planes, with the slowest etch rate at the (111) planes.  If starting with (100) silicon wafers, this leaves a surface of random pyramids with a 54.74 degree tip angle. [9]  If you watch the movie in this reference [8] , notice that while the operator is wearing gloves, he is not wearing a "bunny suit" like a modern integrated circuit (IC) fab.  The cleanliness of a PV cell line is much, much less stringent than an IC fab.   High efficiency cells chemically or mechanically polish the back side, either at this step or later.  This makes the back side more efficient, since a larger surface area increases a loss mechanism known as surface recombination.  Basically repeat step 1 with a lower concentration of sodium hydroxide at lower temperature, and add some isopropanol. A cover or reflux condenser will cut your losses of IPA.

Multicrystalline silicon cannot be well textured this way, since the crystals are at random orientations.  Typically a mixture of hydrofluoric and nitric acids is used.  The texture isn't the best, but it's better than doing nothing.

Step 3 is to form the emitter.  This is done by diffusing an n-type dopant, phosphorus, into the front of the cell.  There are two main methods, POCL (phosphorus oxy chloride) or  phosphoric acid.  POCL, chemical formula POCl3, is a volatile liquid that is introduced into a closed furnace via a stream of inert gas, along with oxygen, at around 700 or 800 deg. C. [10]  The furnace will have an electrically heated fused silica tube with the wafers in slots in "boats" of fused silica.   After making a coating of phosphorous "glass" on the wafers, the POCL is turned off, and the temperature raised for a while to "drive in" the dopant.  Typical depth of an emitter made this way is around 1 um.  While this is a batch process and thus less suited to mass production, it is very repeatable and clean.

The other method is to spray a solution of phosphoric acid onto wafers using either an ultrasonic or gas atomization spray. [22]  Some kind of wetting agent, like butanol is used to keep the phosphoric acid from balling up before it dries. [23]  Butanol is a product of the acetone producing fermentation mentioned above.  Then the wafers are typically fed into a belt (or ceramic roller) furnace to "drive in" the dopant at up to 1000 deg. C. [11]  This has the advantage of being a continuous process, but it's not as clean and stable.  At home you can use an electric kiln for this if you don't mind some attack on the heating elements.  No shielding gas is needed, in fact a bit of oxygen from the air is helpful.

Step 4 is the post diffusion PSG (PhosphoSilicate Glass) etch.  The layer of PSG formed during diffusion prevents contact being made with the cell, and particularly for multi crystalline cells, has drawn out a lot of contaminates.  Hydrofluoric acid in fairly dilute concentration is used.  Back to your hotplate and plastic container.  The waste can be neutralized with the waste from step's 1 and 2.

Step 5 is edge isolation, also known as emitter isolation.  In the two main processes for emitter formation described above, the emitter wraps all over the solar cell, electrically shorting the front to the back.  In the old days, a plasma etch was done on a stack of cells, using fluoride containing gases in essentially a big microwave oven.  The gases are bad for the ozone layer and global warming, and the stacking/unstacking damaged a lot of cells.  Then people started using lasers to trim a line around the edge of the wafer on one side or the other.  Now, single-sided wet processes have been developed that combine steps 4 and 5 into one.  The cells are put onto rollers that keeps only one side wetted by an etch solution.  Typically the back is etched and polished using HF and nitric acid, then the wafers flipped into another line which etches the PSG using plain HF.  [12]

At home I think I'd try a tray with a little stand in it.  Set the wafer backside down on the stand.  Use a syringe/pipette to fill the tray up to the bottom of the wafer.  An L-shaped rod would help wet the whole back and keep the HF/nitric acid mixed/circulating.  A bit of air blowing down on the wafer will keep unwanted etching from the front.  When the back is done (via time), take the wafer out, rinse, then toss in a breaker of dilute HF for a time you've empirically determined is long enough.  Remove and rinse and let your new solar cell (albeit without contacts) dry.

Step 6 is typically the deposition of an Anti-Reflective Coating (ARC).  This days it is mostly silicon nitride (SiNx), deposited with plasma enhanced chemical vapor deposition (PECVD). Besides enhancing the light trapping of the solar cell, the SiNx reduced surface recombination losses by the fixed charge in the SiNx, and also trapping some hydrogen from the deposition process which latches onto dangling atomic bonds of the silicon surface, passivating them. [13]

The feedstocks are silane gas and ammonia.  In part 1 I've already described one means of making silane, which can also be derived from trichlorosilane by transforming some of it to silane and the rest to silicon tetrachloride.   This reaction takes place under fairly mild conditions as long as one has a decent metal halide catalyst.

Since the ARC is very thin, less than a micron, not much ammonia is needed.  It may be possible to collect enough urine or other organic source.  If that is not possible, the Haber process could be used, though the pressures are rather high (several thousand psi/150-250 bar).  All one needs is hydrogen, nitrogen, the equipment to pressurize and heat them, and the iron based catalyst. [14]

The PECVD machine is a closed chamber, a batch of wafers goes on a tray in the bottom.  The chamber is pumped down to a vacuum, then a tiny bit of silane and ammonia are allowed in while pumping continues.  An electrical plasma is started to disassociated the gases, turning them into ions and radicals, which in turn deposit on the top side of the wafers.  The power requirements depend on the frequency, but for a batch of 16 wafers is fairly modest, a kW or two.  Frequencies used vary from 50 kHz to 10's of MHz.  Exhaust gases need to be burned, resulting in nitrogen, water and silica that is scrubbed out in water.

If all this is too much, one can grow a layer of silicon oxide with steam or with added silane with just a hot sealed furnace.  It's not the best, but it would be better than nothing, particularly for passivation.

A better (than silicon oxide) antireflective coating would be to use titanium dioxide, as they did in the early 1980s.  One starts with titanium ores (oxides), then reduces them with carbon and chlorine.  The resultant titanium tetrachloride is purified by distillation (it boils at 136 C, 278 F) in the absence of water.  Then it's mixed with isopropyl alcohol to get titanium isopropoxide.  The hydrochloric acid coproduct is certainly useful.  Then one may form the ARC in several ways: spray pyrolysis, screen print a thickened layer then fire it, or spin coat a sol gel.  The latter could use some of the hydrochloric acid as a catalyst. [15]

Step 7 is to "metallize the cell" - put on electrical contacts.  Typically this is done with
screen printing of pastes made from silver particles and glass dusts in a vehicle of various organic things.  There are three separate printings on most cells: [17]
        Ag +glass paste = front gridlines and busbars
        Aluminum paste = rear back surface field to increase efficiency [16]
        Ag paste = rear solderable contacts to back.

The screen printing is just what it says it is, a mesh screen with a pattern that blocks the paste from unwanted areas, and a squeegee that forces paste through the openings, just like making a screen printed T-shirt.  Between putting down each layer, the paste must be dried.

For any surface an ARC was put on (some processes can do it one-sided), the pastes will need the glass particles.  The glass fuses during the firing and dissolves the ARC, allowing the silver or aluminum to make contact with the silicon.   Note that Hall/Heroult introduced their aluminum reduction processes in 1888/1889.

If using TiO2 ARC, one probably waits until after metallization.

One could also do mask defined sputtering or evaporation to put down metal on a cell.

At home, for real simple, one could just paint on some paste.  If one didn't have some nice paste with the glass frit, then make some pure silver particles (grinding or silver nitrate) and use that in some alcohol/etc. as paste before any ARC, then fuse in the furnace at around 950 C.  A reducing atmosphere would be nice here, some hydrogen mixed with mostly nitrogen.  Then I might let in a bit of steam to make a high temperature oxide for surface passivation.  These gas steps are optional.

The back surface field is optional, but nice to have for higher efficiency.  An alternative would be to evaporate or sputter aluminum metal onto the back if one had access to such (a chamber, vacuum pump, argon/power supply/aluminum target or heater/aluminum).  The lack of the back surface field will cost you a percent or two in absolute efficiency.

Step 8 is firing the cells at 800-900 C for a minute or so, which fuses the glass and metal particles and creates contact with the cell. [18]

There, you're done with the cell manufacturing.

Cell Testing

Cells are then tested.  These days with artificial light on automated machines.  They could be testing during the daytime against a known reference cell.  Based on the results, they are then sorted into groups with matching power output.  A string of cells will only produce as much current as its worst cell, so it's a waste to build a string of mis-matched cells. [19]

Module Assembly

From here on out, things are fairly low-tech, or at least they can be.  While most of these steps are automated today, it was only a few years ago that it was all done by hand.  The first thing is to solder tinned copper tabs to the good cells. [20]  These will go from the front of one cell to the back of the next.  An ordinary electric soldering iron and plain old solder work just fine. 

Then a string strings of cells is "laid up" in a line, and the tabs from one soldered to the next in line.  A cell gives somewhere between .6 and .7 volts, so one needs about 4 cells for a minimally useful module.  The full sized modules have strings of about 9 cells, and join 4 strings, giving roughly 20 volts output at no load, dropping some under load.  This is a convenient voltage to charge 12 volt batteries with.

After testing the string, some number of strings are laid up, typically on top of the front glass and a layer of encapsulant material, and soldered together by copper ribbons interlayered with more encapsulant material.  Then another sheet of encapsulant goes over the whole thing, then either a plastic backsheet or another piece of glass, then the whole things goes to a laminator.  In the laminator, there is a membrane above the module assembly, and the whole thing is inside a metal box that is now sealed.  A vacuum pump pumps the air out of both the top and bottom compartments, the assembly is heated so the encapsulant melts, then air is let back in the top, so everything is squished together.

After the lamination, one will then attach either a junction box and/or some wire leads to the ends of the ribbons left outside the encapsulant.  Typically these boxes are made of molded plastic, but I'm thinking they might just be dispensed with.  Wire insulation is another issue (is every thing in the modern world made of plastic?), it has to deal with years of outdoor exposure.

There - now you have a solar module, ready to be tested.

The polymer based encapsulants (and backsheet) are the rub.  The most common today is Ethylene-Vinyl Acetate (EVA).  Currently in widespread use for a great many plastic products, it is now made from petroleum.  Ethylene can be generated from ethanol from fermentation.  The vinyl acetate is made one of two ways: ethylene + acetic acid + oxygen, or acetic acid + acetylene.  Acetic acid is what vinegar is made of, by fermentation of ethanol by acetic acid bacteria.  So we have to make a lot of booze to make PV I guess.

Many early solar modules used polydimethylsiloxane (PDMS)  based silicones, but those were/are expensive compared to EVA.  Synthesizing PDMS would involve getting some methanol, likely via the destructive distillation of wood - we need the charcoal anyhow.  The methanol gets mixed with hydrochloric acid to make methyl chloride, which gets reacted with some of the metallurgical grade silicon from our electric arc furnace to make dimethyldichlorosilane (among other things).  After some separation via distillation, we can then figure out how we want to cure it.

A site specific analysis of the economic factors involved would be necessary before making a decision on which way to go.  Both EVA (at least the well-made stuff) and silicone encapsulated modules have lasted 25+ years.

If either polymer (or alternatives) simply was not easily available, then you could make a module by mounting the cells on an insulated metal sheet pan, and bolting that to a sheet of glass.  It need not be completely sealed to take care of thermal expansion of the gases inside, though a breather plug to keep the bugs and dust out would be good.  Outgassing of interior components should be avoided, a fibrous mineral insulation could be used, or blown glass insulators strung together on metal wire.  There are many ways to skin the PV cat.

Another issue is whether to use a polymer backsheet or another sheet of glass.  Unfortunately, the only polymers that hang around long enough for a reasonable module life (25 - 30 years) are made of fluorinated plastics, and are also produced on fairly large scale on rather sophisticated calendering/laminating machines.   I'm thinking another sheet of glass, that we don't have to bother making low iron, will be the way to go.  Rather than building a float glass plant and a backsheet plant, just use the float glass plant more.

Discussion

Looking in back through the steps I don't find any complete show stoppers.  Sulphuric acid to make the other acids seems easy enough.  Sodium hydroxide is easy as long as we have electricity and salt water.  The scary one was looking for plastic containers to do the wet processing in, and we'll either salvage that or do ethanol based polyethylene.  The SiNx ARC is the hardest part I think, for two reasons.  One is the RF plasma generator, though there are successful systems using both microwaves and lower frequency (50 kHz), at a few hundred watts, and that ought to be fairly easy.  The other is the ammonia, which is a big undertaking to get the combination of pressure and temperature.

Concerns about the lack of sophisticated semiconductor fab are misplaced, the first PV cells were made by guys dressed in wool suits.  If you look at picture #4 in this reference, you see the early diffusion ovens, even for transistors, were basically overgrown kitchen ovens.  The women operators aren't even wearing hair nets. [21]

Encapsulants will be the big issue, which will affect any PV technology.  If we assume a 12% efficient (poor for now, but way better than nothing) module, down due to the more crude nature of things, and a .5 mm layer of EVA, for 1 MWp we need 8,334 square meters x 2 sides x .0005 meter thick = roughly 8.4 tonnes of EVA assuming no waste.  Thats are fair bit of booze to come up with without industrial farming.

An open issue are the UV protectants for the EVA, and the curing catalysts for both the EVA and silicones.  Some of these chemical are fairly exotic.  And there's a lot of empirical information that is often proprietary to the companies that now produce the polymers, which means that in a more rapid/disruptive collapse that re-inventing the needed subset of the plastics industry will not go so smoothly.

CIGS, CZTS and CdTe might be worth looking at - it would take some real detailed study to see what would really be sustainable.  But for now the market is voting crystalline silicon.

Conclusion

Again, I can't really conclude without knowing more about the environment the proposed operation would find itself in.  The current existence of many 10's of GWp/year production capacity says things are possible, but the difficulties of reduced transportation and less access to petrochemicals are open issues.  Turning silicon wafers into PV is not a one-person operation if one had to make everything by oneself, but gets rather involved with a lot of parts and chemicals.  But neither is it a priori impossible to assemble a team of dozens of workers at small but usable scale.

==========
references
[1] http://www.cnn.com/2012/01/10/tech/innovation/solar-powered-led-lamps/

[2] http://deciwatt.org/

[3] http://light.lbl.gov/pubs/tr/Lumina-TR10-health-impacts.pdf
more info on low-carbon off-grid light for the developing world:
http://light.lbl.gov/pubs.html

[4] http://sundanzer.com

[5] http://www.treehugger.com/clean-technology/new-solar-refrigerator-prototype-from-chile.html

[6] http://opensourceecology.org/wiki/Polyethylene_from_Ethanol
http://www.braskem.com.br/site.aspx/plastic-green

[7]  http://en.wikipedia.org/wiki/Acetone-butanol-ethanol_fermentation

[8] http://www.pveducation.org/pvcdrom/manufacturing/texturing

[9] http://memslibrary.com/guest-articles/47-silicon-etching/33-anisotropic-wet-etching-of-silicon-with-alkaline-etchants.html

[10] http://sun.anu.edu.au/files/ltbfp_PhosDiffn.pdf

[11] http://www.pveducation.org/pvcdrom/manufacturing/emitter-diffusion

[12] http://www.ise.fraunhofer.de/de/veroeffentlichungen/konferenzbeitraege/2008/23th-european-photovoltaic-solar-energy-conference-valencia-spain-2008/single-side-etching-key-technology-for-industrial-high-efficiency-processing

http://www.rena.com/fileadmin/img/Produkte/200_Solartechnik/220_Inline/222_InOx/RENA_DB_InOxSide_20110829_final2.pdf

[13] http://pveducation.org/pvcdrom/design/anti-reflection-coatings

http://pveducation.org/pvcdrom/manufacturing/anti-reflection-coatings

[14] en.wikipedia.org/wiki/Haber_process

[15] http://www.journalamme.org/papers_vol52_1/5211.pdf

[16] http://pveducation.org/pvcdrom/design/surface-recombination

[17] http://pveducation.org/pvcdrom/manufacturing/screen-print-front

[18] http://pveducation.org/pvcdrom/manufacturing/firing

[19] http://pveducation.org/pvcdrom/modules/mismatch-for-cells-connected-in-series

[20] http://pveducation.org/pvcdrom/manufacturing/module

[21] http://www.computerhistory.org/semiconductor/timeline/1954-Diffusion.html

[22] http://www.sono-tek.com/chemcoat-silicon-doping

[23] http://iopscience.iop.org/0022-3727/40/9/008

[24] http://www.seia.org/research-resources/us-solar-market-insight-2012-year-review

[25] http://www.greentechmedia.com/articles/read/Germany-Hits-59-Renewable-Peak-Grid-Does-Not-Explode

[26] Tracking the Sun is now put out yearly by the Lawrence Berkeley national lab
http://emp.lbl.gov/sites/all/files/lbnl-6350e.pdf

[27] http://www.renewableenergyworld.com/rea/news/article/2013/02/100-gw-of-solar-pv-now-installed-in-the-world-today

====
edits
20140118 - some typos about removing the PSG, fixed factual error that tube radios require 10's of watts due to existence proof of Russian thermo-electric generator powered radio at:
http://www.douglas-self.com/MUSEUM/POWER/thermoelectric/thermoelectric.htm#rl
90 V * 0.012 A = 1.08 W,  + 1.5 V * 0.25 A = 0.375 W = 1.455 W
Added 100 GWp of PV cumulative install.