- Bypass on BOTH VCC and GND?  :/  Not gonna.

x Read every component datasheet top to bottom one more time...
	x What about the PCB 'safe' zone under the USB mini-B?  DONE

x Re-place your capacitors -- min 35 mils pad to pad
	x Kemet shows vias pointing down, as close together near the center of the pad as possible -- GOOD IDEA!
	x At some point I decided 35 mils from pads was okay, so I'm sticking to it...  Need minimum nH!
	x Fairchild recommends pin->cap->via approach with 100 mil max trace
	x AMCC recommends the bulk decoupling capacitor is placed within 300 mils of the PCB power entry point
		Lots of people are recommending bulk decoupling be close to the power 'source'
	x Altera recommends that the VCC bulk decoupling capacitor be placed right at the regulator output
	- Altera recommends a ferrite bead _at_the_PCB_source_ (before the decoup capacitor)
		This could prevent our noise from tripping up the Jag and vice-versa
		Scratch that -- our switching power supply should keep VCC clean, and GND is dirty anyway
	- AMCC recommends capacitors are located within 150 mils of the GROUND PIN (not VCC)
		We have some ground pins with no capacitors that close on EZ-HOST!
	x The AMCC uses the 'bad' pin->cap->via approach -- actually, they route ground through decoup caps too
	x Xilinx recommends pad->via traces be equal in width to the via (i.e., 30mil) and as close as 30 mils
	- YUCK:  Bypass capacitors generate big EMI fields that disrupt nearby clock and USB traces

	x Lots of people have said you need bypass caps near layer changes on fast signals
		x One example is a bypass cap near a USB mini-AB 'peninsula' -- due to probable vias
		x Most people suggest it is okay to 'borrow' a nearby power cap for this -- and we have many

x Smaller bulk caps?  Yes, but... stick with tantalum!
	Ceramic bulk caps are getting more popular... and they are smaller and as cheap as tantalum for 10uF
	Kemet says:  "Large cap value ceramics may provide effective decoupling... at 1/4th tantalum uF"
		They say the biggest risk is temperature sensitivity -- especially in Y5/Z5, but even in X7R!
		Y5R - 82% capacitance loss over -30+85C, where X5R is 15% over -30+85C, and X7R 15%/-55+125C
		X7R hold up _way_ better with DC voltage -- 16V rated @ 5V is -20% X7R versus -60% Y5R
	x Most app notes suggest 10uF is enough for my application -- I don't why I'm pushing 47...
		If you choose ceramics, go with 10uF -- 3x or 4x overrated
		The TPS-2149 only wants 4.7uF on the input, and 'at least' 4.7uF on the LDO output...
		They also need ESR between 200 mO and 10 O -- they give ceramic a blessing right there!
	x SAY NO TO CERAMICS:  "Circuit stability can actually suffer if you try to do all of your
				power supply bypassing using low-ESR, high-Q capacitors."
		The EMI doctor also says basically the same thing -- low-Q is your friend to swamp resonance
	x Good small Tantalums (3216-18 is a little bigger than an 0805):
		T491A106M016AT	10uF/16V/3216-18	$0.18/100
		B45196H3106K109 10uF/16V/3216-18	$0.16/100
	x These are cheaper than our $0.49/100 Tants...

x Ongoing TPS-2149 confusion...
	x UPDATE:  Found local VDO decoups on all appnotes, implemented that, not on OUT1
	x The appnote in the datasheet shows 0.1uF decoupling on VDO, not OUT1
	x The description doesn't say anything about VDO, but suggests you want it on OUT1 to avoid shorts
	x A separate appnote shows 22 pF on OUT1, 0.1uF on VIN, bulk on OUT1 and VIN (100uF!) and nothing on VDO
	x My conclusion:
		- Make sure Flash decouple is right next to VDO to cover that base
		x Decouple OUT1 with 0.1uF + 150uF aluminum
		x Decouple VIN with 0.1uF + 10uF ceramic (near PCB port of entry)
			x The logic here is that most bulk power is in Jaguar power supply -- not needed here
			x You should not need a ferrite bead since noise is not likely to escape on VCC5
		x Decouple VDO with 10uF ceramic (right at output)

x Clock layout
	x Do not run traces under clock
	- Move clock capacitors as far from EZ-HOST ground returns as possible
	- Keep clock as far from bypass capacitors as possible -- oops

x USB trace layout
	x Even nearby power traces will cross-couple enough with one side to modify the pair impedence
		Most people say you need 3X - 5X spacing on either side
	- Everybody suggests USB traces belong adjacent to GND plane, just like clock traces
		Some people put GND plane on layer 3 when they have surface mount components, for this reason!
	- Probably need a local pull-up on OVRCURL if we are going to use it at all...
		It is just too ugly to run such a noisy trace _right_next_to_ the USB signal lines
	- VBUS is probably even worse -- it has a 2K pull-down in EZ-HOST, but has gotta be noisy...
		At least the nearby lines don't need to work until VBUS is being held high?
		The app notes say no resistors are required, but the CY4640 provides a no-load pull-down
		The app notes suggest transient protection is not required on OTG VBUS
	x More spacing has been added...

x Edge connector assembly
	x Which carts are gold plated?  All of them.
	x What are the risks of not doing this?
		Tin plating can corrode when it touches the gold connector  ACK
		Copper will quickly tear up the connector and itself during swap copy
	x Immersion (soft) gold versus electrolytic (hard) gold?
		Only hard gold is suitable for edge connectors...
	- Need special quote from eTekNet on hard gold...
	- Need to route the edge connector at 20 degrees according to Atari (many people use 30)

x EMI prevention
	x USB shield should be tied to ground with a decoupling cap to keep it from resonating too much
		x Most people suggest 0.1 uF -- apparently it works more like an inductor anyway
		- One guy on eng-tips makes fun of the capacitor idea and says just use a fancy trace
		- One Cypress note says use 'some resistor value -- you figure it out'
		x Another Cypress note says to use a 1M resistor plus a decoup capacitor (4700pF/250V)
			This probably prevents ESD from ruining your day
		- Many people suggest (including antidotes) that powered devices should NEVER couple shield to GND
			- Apparently hosts usually do, and this leads to a device overheating due to ground loop
		AC coupling (capacitor) provides the necessary immunity for the device port
		DC coupling (ferrite bead) works best for host ports
	- Decoupling capacitors should be low ESR!
	- Pull-ups should be near their open-collector outputs -- we snake OVRCURL about 4 inches  :-/
	x A lot of people say avoid ferrite beads on GND because it is too critical -- can't take fluctuation

o ESD protection
	o Don't butt the USB connectors against the surface -- leave some plastic clearance like your laptop

x What capacitor for VPP?  Probably a multilayer ceramic 0603 or 0805?
	- You might be thinking of capacitor arrays, not multilayer ceramics?  Nope!  Don't want that
	- We need a separate .1 uF high frequency decoupling capacitor _and_ bulk capacitor...
	- One DC/DC app note states that VPP capacitance can be in the 2uF range

	ECJ-1VB1A225M		2.2uF/10V - 0603/X5R			- $0.09/100
	LMK212BJ475KG-TR	4.7uF/10V - 0805/X7R			- $0.17/100
	LMK212BJ475KG-T		4.7uF/10V - 0805/X5R			- $0.15/100
	EMK107F225ZA-T		2.2uF/10V - 0603/Y5V			- $0.06/100
	LMK212F106ZG-T		10uF/10V  - 0805/Y5V			- $0.16/100
	GRM188F51A475ZE20D	4.7uF/10V - 0603/Y5V 			- $0.11/100
	C2012Y5V1A475Z	  	4.7uF/10V - 0805/Y5V			- $0.10/100
	CC0805ZKY5V6BB335 	3.3uF/10V - 0805/Y5V = bad toler/temp 	- $0.12/100
	C0805C225K8RACTU  	2.2uF/10V - 0805/X7R 			- $0.21/100
	
	x You can get by with 1uF because the SRF is ~9MHz and we are still on the chart at 5.3MHz
	x 1uF 0603 or 0805 gives us 'okay' decoupling and 'okay' bulk capacitance...

x Biggest EMI mistakes right now:
	x Shield ground is not tied to real ground (usually via resistor) -- this is in Cypress usblayout.pdf
	x Link port VBUS is routed right next to Link port D+/D- -- noisy power from host = noise on USB bus
	- USB lines are not routed adjacent to GND plane

x USB 2.0 standard says we should monitor VBUS on our USB peripheral port
 	This is to avoid sending current on DP/DM into a host that is powered down (logical enough)
	Pin 11 is a dedicated VBUS detect pin...  We should look at wiring that up

- We can probably specify C44 No-Load, then add it later -- there is enough bulk decoup in the Jag, right?
	x Too risky for me

- The shield area screws up parsing of the entire Gerber file in GC-Prevue...
	x viewMaster can fix?

x Cross-check your land pad with land pad viewer...  (Found mistakes too)

x USB traces are 12 mils apart when routed diagonally?  WTF?
	x Fact of routing life

x PowerPad hell:
	x The PowerPad method in the IPS 2149 requires vias in pad -- assemblers hate this
		TI recommends plugging the bottom of the vias using solder mask to keep solder from escaping
		But Screaming Circuits says that gasses form during reflow that pop the plugs
	x On larger pads, people use solder mask around the vias -- this works just fine!
		But it probably reduces the cooling efficiency...
		-> Also, this requires a good solder mask stencil to keep solder out of all those little 'holes'
	x You're probably better trading off good cooling efficiency for vias well off the pad...
	x DO NOT PUT ANY UNRELATED VIAS UNDER A POWERPAD
		It looks like cheap (silver) PCBs have a chemical reaction that can cause shorts to the PowerPad
	x DO NOT USE WEB OR SPOKE VIAS

- Ensure none of the other chips I'm working with have metal on the bottom... otherwise our vias are screwed

x I'm not using multiple vias yet -- how many and what size are recommended for major power lines?
	x Even 14 mil vias (smaller than standard) are fine for up to 200 ma...  Put two on the 5V line
	x We should tie more ground pins together with single vias

- How should power delivery decoupling look?  My C41 via is kind of stupid looking...

- Conclusion:  Can't do these wiring 'improvements' because they add two resistors AND DON'T WORK:
	x We don't need GPIO as a register 'back-door'.  We can write to 8xxx to disable the ROM image
		x At 9 48MHz cycles per 5.319MHz bus cycle, we can handle FIFO and register I/O in 1 data port
		x PWM is probably outputting 8MHz -- this is simplest/safest -- I/O on low edge, delay on high
		x Worst case:    WWWXXXAAAXXXAAAXXXRRRXXXAAAXXXWWW
			       F--------C--------R--------F--------
	x Can we use SPI for Microwire/E2DATA?
		SCK slaves on GPIO so it is an input
		MISO writes to E2DATA so it is a safe output
		MOSI reads from D0 so it is a safe input
		We must ground nSSI or it all gets ignored (can't use GPIOL1 -- Microwire CS is active high)
	-> BUT -- we are going to drive GPIO-0 and D0 during EZ boot...
		D0 is easy to handle -- EZ-HOST exits reset at the Jag's whim, so the Jag can stay off the bus
	SCK and nSSI are not so easy -- driving these could be quite destructive...
		It appears that EZ-HOST writes C03A on the external bus during bootup (i.e., D9+D10 driven low)
		It also looks like arena_init_find_mem writes aa55 to the first external address...
			But it probably does this in 8-bit mode only, so D10 is not driven low at this point
		D9+D10 goes low for 166ns within 10us of boot-up (verify with logic analyzer)
			-> This is only safe with two resistors in series
		GPIO-0 is an input during reset... but this DOES NOT WORK because it is very brief
			Confirmed in jerry.net/jclk.net that cfgen only tri-states GPIO-O briefly
	-> FORGET ELECTRICAL, THE LOGIC ISN'T FAST ENOUGH!
		We only have 360 EZ-HOST cycles from the last bit of the address to the expected reply
	x The current approach is 'good enough'...  Can't use serial and EEPROM at once, but... who cares?
		Note that the complexity around 'status bits' and 'start bits' is easier to implement here
		The only crappy part is stealing the 'private bus' to send a completed 'word' (byte really)
		The completed byte goes into the HPI mailbox on the same cycle that the last bit arrives
			This gives us bus time -- since the Jaguar memory cycle > HPI memory cycle
			The EZ-HOST has ~384T to pick up the HPI interrupt -- might not be enough time
			-> Remember, the EZ-HOST is basically idle when not in CD/mass storage mode
			Still, this is an argument for reusing the 'register access' mode to give DMA...

x TPS 2149 open issues:
	x OverCurrent signal is an open-drain output
		Do we need a pull-up to read it via GPIO?
			x Scary idea:  Use the built-in pull-ups on USB 2B to detect OverCurrent
				Why not?  We're not using the pins for anything else...
		It needs software debounce...  Give it at least 100ms of registered highs before prompting
		Note that steady state is not achieved -- due to external caps and hysterisis it will cycle
	x VBUS cap needs ESR between 200 milli-ohm and 10 ohm... does our aluminum electrolytic meet that?
		ESR = DF / (2 * pi * f * C)
		Our 150 uF 10V EMVA100ADA151MF55G has a DF (tangent) of 0.30
		0.30 / (2 * pi * 100 * 0.00015) = 3.2 Ohms?  This is consistent with typical caps!
	x They recommend 100mA total for startup current draw...  Holding EZ-HOST in reset helps this
		We have a max current of 200ma -- this should be enough, but check the datasheets!
		Hopefully our funky VPP wiring will help limit current during programming...
			Programming is very intensive with EZ-HOST and USB bus at full use, Flash burning, etc

x CY4640 has the ability to detect VBUS on USB B...  Do we need that to enumerate as a device?
	Yes, but we can use OTGVBUS without extra components according to the datasheet

EMI experts warn that unconnected copper planes become transmission lines...  Like shield ground?  :(

-> Shield ground must be connected to solder points using web or snake vias, especially vias to internal ground
	Otherwise thermal transfer on PTH components will kill the board  :-(

x How do you prevent shorting out the UART when you use the JTAG cable?  You must put GPIO in tri-state...
	x I AM SO GLAD I FOUND THIS!  AGH!

- Why do the holes from my USB connectors look different in the inner layer view?
	Who cares... they render right.

x Most people put VCC under signals on surface mount boards... but lots recommend GND... why?
	Depends on where your fast signals run...  But both arrangements work okay.

x Typical advice says all ground pins should be via'd separately...

x USB trace spacing must be kept constant at all times -- used 8 mil/8 mil

x Distant from via and pad:
	We are very tight in some places... is this a manufacturing limit?
		-> It's an assembly limit!
	Yes... if you're not careful, the via will suck the solder off the pad like a tiny straw in assembly
	The board manufacturer could care less, but the assembler will hate you forever

CD-ROM compatibility:
	It _looks_ like EINT0 and EINT1 and 9V and RW are all hooked up in the CD schematics
	But it's a little vague there...  It's worth doing a conductivity test on RW at least!

- Thermal relief in SMT:
	The CY4640 has little non-plated holes beneath everything that has a lot of solder
	These could be to allow thermal relief in reflow soldering!  Or maybe not.
	CONCLUSION:  Until somebody explains these to me, I'm leaving them out!

- For assembly, what about RoHS?
	x CONCLUSION:  For prototyping, we will not use RoHS
	What is the risk of not being RoHS compliant?  Are all my parts RoHS ready?
		-> In the EU, non-RoHS is now banned as of 1/1/2006
	RoHS (lead-free) runs at higher temperatures -- so RoHS certified parts withstand higher temps
		-> UNLESS ALL OF YOUR PARTS ARE RoHS, YOU HAVE TO USE LEAD -- OR THE PARTS WILL MELT!

x Don't forget a JTAG stub for the prototype boards...
	It is going to pay for itself very fast -- and is very easy to remove later
	The only alternative is a USB-based JTAG controller via EZ-HOST
		Maybe someday I'll be that talented, but not today

x Cypress recommends the following layout guidelines for EZ-HOST:
	Cypress technote "High-speed USB PCB Layout Recommendations
	90 ohm differential impedance

x Too many unknowns about I2S to finish PCB layout...  So don't use EZ-HOST for that!
	PWMs can generate clock enables plus word select on a 32 on/2 off rotation at any frequency...
		SS is /CS for SPI (Slave Select)
		But it's not clear if it works in master mode... probably not?
	MOSI/MISO -- we want to switch between slave (cart game) AND master (CD game)
		But MOSI and MISO are miswired in one of the two modes -- since Jerry just has one wiring
		Can we use 3wire mode in cart game to avoid swapping wiring?
			Probably not -- that probably just tri-states
	x Only solution -- screw EZ-HOST SPI and implement I2S in CPLD

x Is it safe to tie lines directly to ground?
	Most people use resistors to tie to CMOS chips to ground
	I am looking at grounding several pins in the netlist...
		Including some chip selects, WPs, etc
	Do we need pull-downs or a spare pin on a CPLD to do this?

	Google Groups:
		"Older technologies didn't like having their inputs tied directly to the power rail."
	Several posts agree that tying direct to ground on CMOS is 'fine'.  Some say:
		"Pull the inputs up or down through resistors.  That way if you need a spare gate..."
	One guy says:
		"Typically you don't directly tie anything to any voltage rail, if just for current limiting..."
	And somebody counters:
		"Since the input impedance of a CMOS input is somewhere in the XXX megohms, I seriously 
		doubt that a reasonable external resistor is going to make much difference."

	-> Many many people say it is safe, even in the presence of severe ground bounce, to tie to ground!
		Some people also say you might regret tying direct to ground if you need to hack the board
		So they use resistors or clever pad and trace arrangements that are easily cut	

	EZ-HOST schematic ties 'reserved' directly to ground and lets outputs float

x Internal pull-ups on private bus:
	Our private bus requires pull-ups when nobody is driving it...
	Can we use the CPLD to generate these pull-ups?  Or can the CPLD just drive 0s?
	-> CPLD has 'bus-hold' which probably supplies the needed pull-ups

- RD strobe in Jag:
	Confirm that RD strobes on blit so that A1 is not required in CPLD
	No more RD strobe... we are going to rely on Jag cycle counting

x Cap size:
	0603s are tiny -- and 0402s are ridiculous
	0805s are supposedly better for board rework

	We should order new ferrite beads from Jameco for prototypes (not 0402!)
	We should get a high end tantalum for 9V power...

Drill sizes:
	FreePCB lists the finish size -- some board houses expect drill size

SMT support:
	FreePCB makes SMT pads using a polygon fill in the Gerber
	Apparently the software that BatchPCB uses doesn't like this
	(I've found talk on here of other problems which seem related to this)

Bulk caps:
	We have a bulk cap on 5V and a bulk cap on 3V...
	Don't we need a bulk cap for 9V VPP?  If we want reliable programming, probably!

	9V only has a 10uF bulk cap inside the Jag, near the audio converter...
	Heck, 5V has a bigger bulk cap than that!

	One recommendation for VPP decoupling in a DC/DC converter:
		The output capacitor should be chosen on the basis of its
		equivalent series resistance (ESR) and capacitance value.
		Low ESR tantalum surface-mountable capacitors such as
		those made by AVX are well-suited for DC/DC converter
		applications. Inexpensive aluminum electrolytics may have
		excessive ESR, resulting in high output ripple. These
		should be avoided.

	Since we have a DC/DC converter for 3.3V, we need a good output tantalum...
	But, theoretically we just need decoupling on the input, since the Jag has big caps

	A bulk cap on the input should help smoothe things out too...

	All that's left is the 9V bulk cap...
		We can use our 16V decouple ceramic cap no problem here...
		But what about bulk?  Maybe it's time to give up on that.

Ferrite beads:
	EZ-HOST has 1 $0.04 bead on each V-BUS...  Sounds like cheap insurance!
	Intel and TPS2149 notes recommend a bead on ground as well, to reduce EMI?  Hm...
	Research more!

	Do we need anything to protect us against hot plug in our power supply?

x nReset on EZ-HOST -- if driven via CPLD, can we trust unprogrammed?
	Unprogrammed CPLD pulls nReset up -- so there is no POR...
	Maybe we are better off doing initial CPLD program via jumper cable?
	Reset must be routed into TSOP as well, so WP bits are cleared...
	x Need to make a hole in the solder mask 

x The great cap debate:

	Tantalums explode -- derate them if cheap/possible... 3x peak V is good
		Ceramics are the only option for decoupling/bypass
		Tantalums are used for 'bulk' decoupling (large values, low khz)
		But some people say they can be 'bad news across a power rail'
		Tantalums fail into a short circuit -- blows the fuse, bricks the Jag
		DO NOT USE WITHOUT RESISTORS (they say)
	Electrolytics have terrible life...
		And poor ESR -- Tantalums are better for bulk decoup
	High value ceramics are cheap -- like 11 cents/each for our 10uF bulk caps
		But they are narrower bandwidth

	-> USE Electrolytics for USB power drive...
		They are cheaper, but bigger too...  Careful placement is the rule
		And their short lifetime may matter less for USB power delivery...
		Everybody else does it this way...

	Decoupling advice says run power to cap pad, then cap pad to VCC
	I seem to be doing it wrong in places... Power to VCC, then cap pad...
	-> This advice is repeated everywhere -- gotta find a way to do it
	Not everybody likes caps-on-bottom -- though they are an easy way to do this

	They warn about DC signal lines -- all lines tied to VCC I'm guessing

	The TPS2149 app note uses tantalums for bulk decoupling without resistors...

	Actually, the EZ-HOST uses tantalums for all bulk decoupling

	They should have better long term reliability with the occasional explosion...

x Benefits of 4 layers:
	Smaller PCBs -- can fit 18 vs 15
		18 is about as small as we can go -- assuming 33mm high boards

	Cost difference:
		21 panels @ 21 @ 4L = 408 + 11 * 34 + 50 + 21 * 16 = $1168/441 2.65
		24 panels @ 18 @ 4L = 408 + 14 * 34 + 50 + 24 * 16 = $1318/432 3.05
		29 panels @ 15 @ 2L = 238 + 19 * 17 + 50 + 29 * 16 = $1075/435 2.47
		(This assumes CNC mill + dual soldermask + >2000 hole fee)

	Dual layers cost 'only' 58 cents per part (18 vs 15)!  Or maybe less...

	This is totally worth it...
		Routing becomes a cakewalk and the PCB can be very dense
		EMI and correct bypassing is taken care of
		Makes it easier to declare FCC conformity...

	Note that we are likely to exceed 2000 holes per board... 
		You need < 110 vias per part -- seems tough to hit
		We have at least 50 signals coming up from bottom of board...

	If we exceed 2000 holes we've added at least 45 cents/part (@ 18/board)...
		We can save 32 holes by granting direct EZ-HOST bus access
		That screws up the DMA window locking feature -- no good

	Naw, just let the holes overflow...  Maybe we can get 21/board!
		Top-route D0-D7 and bottom-route D8-D15 for crazy density...

x One way to save on holes is to mount the electronics 'backwards':
	Data/address lines are easy to route -- so put them on the chip side
	Control lines are harder to route -- so more backside space is an advantage

	This approach also adds more clearance for bigger USB dongles

	But the USB B port is .431 inches high and we have .125 inches clearance!
		We would have to extend the cart outward in back to fit that...

	And the front side only has .250 inches of clearance... can't grow that!
		The USB A ports need .275 inches of clearance...

	An extra 0.3 inches overhang means A ports are close to centered
		We have 0.250 clearance + 0.275 port + 0.225 space + edge of cart

	So we absolutely HAVE to mount backwards...
		And the back side needs to stick out .3 inches

	This also makes sense in some small way since the dongle labels point out...

x Super-compact edition:
	The back-side bulge begins at the base of the front-side label?
		This means we have a Level III mold -- not a flat parting line

	-> YOU CAN'T DO THAT!
		The console doesn't have room until about .25" under Atari logo
		That sucks because these sockets are like .55" long

	WE NEED A MINI-B PLUG (like $0.48/each for Molex pure SMT style)
		These are only 4mm/.157 inches tall

	Of course, shorter/SMT style plugs need a longer PCB...
		You could also drop the plastic on that side -- but it's cheesy
		The top of the SMT mini-B plug _barely_ reaches the top of the Jag

	So we end up with the mini-B in the center and As on the sides...
		The center dips down flush with the top of the console for mini-B
		Two A-posts to either side give you minimal grip to insert/remove

	The front piece has the usual Jaguar sides and top
		On the inside back we have our logo (in plastic)
		And labeled grooves for the three ports (in plastic)
			HOST, LINK, DISK

	Usual cartridge depth is about .675 inches
		We only have to 'bulge out' to about 0.8 inches on the back
		There is already a slight bulge on the cart there -- just grow it

	-> We need to make sure our max component clearance is < 0.2 inches

x 0603 vs 0402:
	Most people suggest 0603 -- the via area uses up most of the space anyway

x Are tantalums safe?
	They explode and they are very sensitive to peakiness...
	The EZ-HOST uses electrolytic... maybe for a reason?

x Is Flash programming safe?
	It says all commands have a don't care on the address bus...
	This means the CPLD should be able to safely poll status even though the A bus is going crazy!
	-> Worst case we can force GPU to help with bus polling, so we are safe!

- Full Type B versus Mini-B:
	I'VE TRIED EVERYTHING... IT DOES NOT WORK
		I wasted two hours and realized we need 1.375" clearance
		That puts us at 43mm high and it means very generous

	The cartridge doesn't clear the back of the Jaguar until 1.25" high/31.75mm from bottom of the board
	A full type B connector needs 7mm clearance to the edge of the PCB
		Thus, a 38.75mm/1.525" board to just clear the back of the PCB (pre-plastic)
		Right now, our packed design is 35mm/1.3" tall
	At the dead center of the cart, we have only 0.5"/12.7mm clearance from back of PCB
	A full Type B connector needs 11 mm clearance from back of PCB

	-> At the moment the Type B connector starts at 1.125"
		That doesn't fit at all... there is not a half inch back there -- even then it would be scraping
	
	The big disadvantage of a Full Type B connector is that it doesn't fit!
		And the plastic case might be very curvy/funky to support this kind of clearance

	IT DOESN'T FIT.  FORGET ADVANTAGES.  IT DOESN'T FIT.

	The big advantage is that it can be PTH soldered with the type A connectors after scoring
		So we get manufacturing flexibility

	Some houses may not appreciate the Mini-B... but I don't see how Full-B fits!
		They could try scoring boards before SMT part placement anyway -- then break them apart later?
		Really clever manufacturers might use tab routing, leaving a mouse-bitten tab beneath the mini-B
			Very easy to snap apart then

	Scoring could play into our hands because our PCB edge could use a slope -- V-scoring is double-sided

- Other (crappy) options:
	You could do a mini-B through-hole that pokes straight out from the board...  That could be post-mounted
	You could use a type A connector and just miswire -- other devices do this (boo, hiss)

x Surface mount it is... hopefully scoring+pick-and-place+snapping saves the day

x AT49BV320D vs AT49BV322D -- cost the same, perform the same, sector lock the same
	The only small advantage of the 322D is its 'RDY' line (no longer needed with CPLD)
	The chip erase feature in the 322D is convenient but sector erase lets us clear just 2MB
		This could hasten boot-up for Tempest  :D
