Novice to manufacturing - so many questions!

[charliehorse55]

I'm working on a consumer electronic product that's rather.... large. It's more or less a large array of duplicated circuits. Rough specifications:

* 8 layers (1oz)
* 450x420mm
* 2000 SMT components
* ~25 unique BOM
* Initial production run - 5k units

This is the first product that I've ever tried to bring to market. I've been thinking about a few different options - either have it built entirely in China, or use a factory in my hometown (Toronto, Canada). There is value to having the production local, but I'm not sure if the price increase will be too high to justify it. I'm hoping the large size of the PCB will put more of the cost/unit into materials than labour. I also understand that setting up production in China would probably involve a trip over there, which I'm able to do.


My second problem is component sourcing. Most of the SMT components are 2 channel discrete mosfets, like the one in my prototype. I need over 1,000 of that part alone, per board. The cheapest mosfet I can find from Digikey or Mouser, even in extreme volume, is $0.05. That's $50/board of BOM right there! Compared to this one I found on LCSC. I'd have to assume it'd be even cheaper buying them in the millions straight from the manufacturer.

Some questions:

• How much cheaper would it be to buy components directly from the manufacturer? How indicative are the prices on distributor websites?
• Could it be good to source the components from China, but have the board built here?
• PCBWay's instant quote (with all the details filled out) gave me $60/board in 1000 quantity. Sounded too good to be true. Does this sound right?
• I've heard that it's actually easier to find factories for consumer electronics in China than the US. Can this really be true? Are all the US fabs more focused on higher-cost products?
• Could a North American based Chinese-sourcing company be worth using? Do they typically charge a fixed amount, or take a %?

Finally, if there's anything else I'm not aware of, please do tell! I'm very new.

[jmelson]

PCBWay is the only Chinese fabricator/assembler that I have used.  They will buy parts from Digi-Key or other big distributor, if you want them to.  Otherwise, I would suggest getting samples of anything special other than resistors and capacitors.  PCBWay is VERY good, I don't think you actually need to travel there to get this done, unless there is something really special about your board.  This is their business, and they know it well.
Jon

[mon2]

PCBWay's instant quote (with all the details filled out) gave me $60/board in 1000 quantity.

Is this the bare PCB with your BOM + PCB assembly fees?

We are down the 401 in Windsor, Ontario and have used assorted PCB assembly houses in Toronto / Markham with mixed reviews. Locally, they wish to retire off a single project. Been to HK many times over the past 20+ years and was in Shenzhen this last October for Nepcon, China. China will offer a very discounted PCBA fee with hopes of making a markup on the BOM. Often this results in a misunderstanding of the true PCBA fee. That is, for a quote of say $1.50 - $2.00 USD per board in 1k qty resulted in actually $10-$15 USD per board by the time they included their parts kitting service. Their claim is that they cannot trust the random kit of parts we wish to drop ship to their factory. They do not want to have finger pointing on who to blame if the PCBA does not work as designed.

If using offshore, be ready for some very late hour chats with the factory. They will hold the SMT line for '15 minutes' for your approval else the product will run as per their golden sample - good or bad. Be 1000% sure the product is functional before even considering offshore.

There are many front companies in Asia and fairly sure that PCBWay is one of them. That is, they have other names and the true PCB house is someone else. Most are brokers overseas. Have seen this repeatedly from attending various electronic fairs. Nothing wrong with that as long as you are aware that there is a risk of losing the tooling that you may have purchased. If the middle agent disappears / retires / quits - the tooling will be under their name and not yours. We did get lucky a few times to be able to convince the PCB house (Suntak PCB in this case) that the IP was ours so did not have to pay for fresh tooling.

At the start, do consider local shops to build a solid product. Then go offshore if it makes business sense to you. The customs in China can be anal for no reason other than ' how dare you export goods INTO China - China should be exporting to you ' attitude. We had thousands of HEAVY DSUB connectors we purchased from China -> shipped to Canada -> sent back to China for PCBA -> rejected back to Canada -> then we shipped it back to China using a different courier = HELL and expensive. We lost about 2 months on this dance. Vendor said to just declare the goods for less. So lie about it??? We did not want to do this but were forced to do so. Yup, China made money selling the goods + local taxes and duties were paid + China made money again in importing these parts. So where are the savings?? No savings other than we did not have to deal with the stuck up PCB houses in Canada. We have our own SMT line and other than it takes time, prefer this over offshoring.

[Mangozac]

In the quantities you're talking I'm not aware of any manufacturers who will sell MOSFETs directly to you.

Most western assembled products use at least some components sourced directly from China, so it's pretty normal to do that and then assemble locally.

[Jester]

Just make sure your not putting the cart before the horse. Are you are unquestionably sure your design is sound before going to volume.

What testing have you done on your prototypes?

How many did you test?

Does it meet all the certifications you require?

Have you considered/calculated your manufacturing test plan and the costs associated with it?

Have you determined import taxes, duties and the CDN$

Prices out of China are MUCH lower than GTA, and often quality is higher unless you go to the expensive assemblers in the GTA.

What CM's in the GTA have provided quotes?

[mon2]

Actually - many will sell direct to you in the noted volumes. Met some in Shenzhen / HK last October. Agree that the product best be solid before considering offshoring. Even then, insist on the first few boards off the assembly line to be shipped to you for testing. Yes, it will cost you more money since they are to build a small batch -> then down their machines for another job while you approve the widget.

Do you have a PCB shop in mind for the bare PCB? Even this can be a nightmare unless you are aware of the capabilities of the vendors.

Here is one of them:

Good day to you!
Last time you said diode is too cheap and it seems unnecessary to find a replacement. Maybe you can give us an opportunity to quote, so that you can compare the price with full forecast quantity.
Any reply will be appreciated.

Carmen Zhu (朱娜娜)
Anbon Semi (安邦半導體)
Web: www.anbonsemi.com
Tel: +86-755-2377 6891(Ext.634)
Fax: +86-755-8148 2182   
Mobile/Whatsapp: +86 18902455618
Email: carmen[at]anbonsemi.com
以客戶爲中心，以奮鬥者爲本

[Smokey]

Few thoughts (USA centric, but should apply to Canada):
1) Get a real live person contact at the major distributors like Arrow/Avnet/Digikey.  They will be able to negotiate pricing.  Don't just use the web stores.
2) I've had production bare boards made at PCBway, but never done assembly with them.  For the boards, be aware that production quantities carry additional expenses that may not show up immediately like shipping and import/customs fees which may be charged by the shipper.  I can only imagine how much that would increase for PCB assemblies that are much more bulky and don't stack/ship as compact.  The higher value of the assembled boards will increase import fees as well.  Make sure you do your homework here.
3) I do all my PCB assembly locally (California) with builds around 1000 pcs.  I have a relationship with my CM, they do good work, and I drive over and pick up the finished products.  It all makes sense for my products, markets, and workflows.  I would recommend at least investigating what local relationships might look like first and seeing what that does for your budgeting.  If possible, price your product such that you make acceptable margins with the local manufacturing.
4) You didn't mention regulatory certs but depending on what markets you are selling into, your stuff might get stuck in customs if that stuff isn't in order.  At those volumes, expect someone to actually check CE/FCC/etc.

[charliehorse55]

Thanks for such detailed responses. To answer some questions:

What testing have you done on your prototypes?

The first prototype proved the concept works, I've just sent out the 2nd prototype with refinements. The hardware is feeling close to being done, the software still needs a lot of work.

How many did you test?

So far I've just been ordering 3-5 boards for each prototype.

Does it meet all the certifications you require?

This is a good question, not even sure which certs are needed. The product is a game, so I shouldn't need anything special. Just whatever the minimum is for consumer electronics. As far as EMI goes, I haven't started on it, but I don't think it will be much of an issue. There is a 16 MHz crystal (placed right beside MCUs), a 1 MHz UART bus that connects them, and then a ~500kHz PWM circuit. I understand EMI concerns start about 30MHz+? That's a pretty high harmonic for 1MHz. I will be using a premade single board computer to interface with WiFi/Bluetooth, but they already come with FCC/CE certifications on them, so I assume they're not likely to mess up the tests when integrated along with the rest of my product? Currently looking at using this one.

 

Have you considered/calculated your manufacturing test plan and the costs associated with it?

Luckily the board can sense what it's doing, so it should be possible to get the board to perform it's own self-test. Perhaps a technician might need to look at it when it's finished to verify it reached the final pattern.

 

Have you determined import taxes, duties and the CDN$

Haven't looked into that yet. The product is a moderately high-end game, so it's likely to sell mainly to western countries.  Going to need to explore distribution solutions as well. It seems like it would save quite a lot on shipping to get a pallet of them sent by sea to major continents before sending them out individually with a courier.

 

What CM's in the GTA have provided quotes?

Haven't gotten any yet. The local fab I've been using for prototypes wants to see the absolute final design before giving their quote. I asked for a ballpark, based on the same information I put into PCBWay instant quote, but they wouldn't tell me anything.

Is this the bare PCB with your BOM + PCB assembly fees?

Yes, assembly with customer supplied parts. They quoted $15/board for the assembly. The PCB is 5/5mil spacing, 8 layer, no special processes. I could increase the tolerances further if it would have a significant effect on cost.

Do you have a PCB shop in mind for the bare PCB? Even this can be a nightmare unless you are aware of the capabilities of the vendors.

At this point, I'm kinda leaning towards using the PCBway "OEM/EMS" to handle everything. If the product can already be assembled/packaged, it could be shipped directly from china to NA/europe/wherever on pallets.

Even then, insist on the first few boards off the assembly line to be shipped to you for testing

Definitely going to do this. I remember reading that Honda flies fully assembled vehicles from their North American plants back to Japan weekly to keep an eye on quality.

[Smokey]

Does it meet all the certifications you require?

This is a good question, not even sure which certs are needed. The product is a game, so I shouldn't need anything special.

If you are serious about being legit and not just winging it (...hoping nobody finds out...), you should have a conversation with a compliance company and nail that down. 

USA is probably FCC part 15 (anything with a clock over 9kHz needs testing).  Each sales region has different additional requirements.  So CE for Europe is not included in the FCC stuff.

I went through Part 15 for a ble device, so intentional radiator, and it ended up taking 2 days and cost under 10k.  But I also passed on the first round.  If I had failed it would have been another round of revision and retest. 

Just a heads up, it's the edge rate that fails emi, not the clock speed.  So if you are switching thousands of fets you may want to get a pre-compliance test setup to see if you are in the ballpark.  A spectrum analyzer and near field probes are a start.

[erikfriesen]

I’m not sure why you say you can’t find mosfets for less than $0.05 I see some options for less than .017 single channel in quantity, although perhaps your design has some specs not mentioned.

I wouldn’t build 5000 initially, it’s a risk. Gremlins always show up in the first 100 that don’t show up in prototyping. If you do a run of 100-200 see how sales are, see what kind of repairs you get into, then you’ll be in a much better position for rev 1 higher count production.  Also, definitely test fcc before the production run.

Until a human has laid eyes on your quote in China, I’d take it with quite a bit of salt.  Also, if this is easily reproduced, you might rather keep your ip in the USA.  Do you want to hand repair 5000 pc of some wrong part (never mind whose fault, maybe they put the wrong feeder into some slot)? Or try to send 5000 back for repairs?

There are other aspects, like ramping from 5 to 5000 will have challenges.  I’ve read something like this.. the first few are hard, then it’s easier to about 100+ where you hit another hard stage, then another at 1000+. This includes a lot of angles from assembly to shipping, programming, whatever.

My $0.10

[c64]

Is this the bare PCB with your BOM + PCB assembly fees?

Yes, assembly with customer supplied parts. They quoted $15/board for the assembly. The PCB is 5/5mil spacing, 8 layer, no special processes. I could increase the tolerances further if it would have a significant effect on cost.

Initial production run - 5k units, $15/board for the assembly = $75k.
For this money you can probably buy good PnP machine, oven, etc. and still have some money left?

[Smokey]

Is this the bare PCB with your BOM + PCB assembly fees?

Yes, assembly with customer supplied parts. They quoted $15/board for the assembly. The PCB is 5/5mil spacing, 8 layer, no special processes. I could increase the tolerances further if it would have a significant effect on cost.

Initial production run - 5k units, $15/board for the assembly = $75k.
For this money you can probably buy good PnP machine, oven, etc. and still have some money left?

I'm a broken record here, but don't do your own board assembly unless you want to become a board assembly company too.  Focus on your product and let someone else that has spent the time and capital to execute the level of quality required to consistently assemble boards do their jobs.  Just buying the machines isnt even close to making 5000 boards that all work. 

[bookaboo]

Whatever you so do not build 5k on the initial run.
Don't build anything you don't have orders for, even if you do have orders build in as small batches as is practicable, keep the batches small until you have complete confidence in the design, then still keep the batches small.

Large batch sizes for "cost reduction" are massively over valued. All sorts can go wrong:
- Defect in design causing 5k rework units
- Defect on the line causing 5k rework units
- Customer changes spec while you have delivered 2k of the 5k, who pays?

Even if all goes perfect, you have to pay for and store this stock. manage the inventory, insure it etc.

[SMTech]

Whatever you so do not build 5k on the initial run.
Don't build anything you don't have orders for, even if you do have orders build in as small batches as is practicable, keep the batches small until you have complete confidence in the design, then still keep the batches small.

Large batch sizes for "cost reduction" are massively over valued. All sorts can go wrong:
- Defect in design causing 5k rework units
- Defect on the line causing 5k rework units
- Customer changes spec while you have delivered 2k of the 5k, who pays?

Even if all goes perfect, you have to pay for and store this stock. manage the inventory, insure it etc.

A touch cautious perhaps, but certainly if you are not the project originator don't over build, that would not be a good time. Equally certainly don't go from prototype to 5K, particularly in this case  where a game infers lots of human interaction. Interaction will find all sorts of bugs you might not catch in testing, however well you follow modern development procedures. Some software bugs might turn out to be hardware bugs etc. I'd stay in smaller batches for at least 1K units and evaluate scaling once you are 100%  the design is stable. After that balance your held inventory against distributions expectations of volume and lead time. Accountants hate stock on a  shelf.

[charliehorse55]

Good advice for sure, thanks guys. The "5k initial order" was based on what I believe I can do within 6 months of launching the product.

My plan now is to do some small batch(es) and distribute to friends and family, working out the kinks in hardware, software, and manufacturing. Meanwhile I'll be taking pre-orders, which will allow for informed production sizing.

There's probably some wisdom in keeping the batches relatively small at the start, even if I have a stack of orders + previously successful test batches.

[tszaboo]

Since you are buying 5M FETs, you have quite a bit of negotiating power here. You might be able to set prices yourself, and ask for that. Negotiate, talk with the boring businessman. Payment terms might be more important than the price itself, because if you can get it with NET60, make the boards and ship it to customer and get paid, you don't need any of your own money invested at all.
China might seem better at pricing, but for larger runs with SMD components the price difference can be negligible or better in the west. You have to calculate with import, shipping delays and other factors, not only final price. Honestly, don't do it, But also be aware that you need to shop around in the west for the prices. There are a lot of smaller assembly houses that would be happy to do this for you for similar than the Chinese prices. And many medium ones that will do the sourcing of the components for you, and have a better negotiating power for better prices.

[mikeselectricstuff]

If you need 1000 MOSFETS and an 8 layer PCB for  "consumer game" product I suspect you might be doing something wrong.

[SMTech]

If you need 1000 MOSFETS and an 8 layer PCB for  "consumer game" product I suspect you might be doing something wrong.

That is the other big question, are you sure you have designed this with manufacturing, cost and simplicity in mind? Because as Mike points out, it doesn't sound like it.

[charliehorse55]

If you need 1000 MOSFETS and an 8 layer PCB for  "consumer game" product I suspect you might be doing something wrong.

That is the other big question, are you sure you have designed this with manufacturing, cost and simplicity in mind? Because as Mike points out, it doesn't sound like it.

I have spent quite a while optimizing for cost/simplicity already. The layers are mainly for PCB coils, which need to have fine-grained control, hence the mosfets. By using 2-channel SOT23-6L fets I am saving on both assembly part count and cost. I have the enable circuit down to ~$0.05 in BOM, there are just 350 copies of it...

It's a pretty unique design and I can't wait to announce it sometime next month.

[mikeselectricstuff]

I have spent quite a while optimizing for cost/simplicity already. The layers are mainly for PCB coils, which need to have fine-grained control, hence the mosfets. By using 2-channel SOT23-6L fets I am saving on both assembly part count and cost. I have the enable circuit down to ~$0.05 in BOM, there are just 350 copies of it...

$0.05 plus the cost of assembling those parts, plus test & any rework.
Can you not use multiplexing ?

[aeg]

The layers are mainly for PCB coils, which need to have fine-grained control, hence the mosfets. By using 2-channel SOT23-6L fets I am saving on both assembly part count and cost. I have the enable circuit down to ~$0.05 in BOM, there are just 350 copies of it...

Are you using 1000 FETs to drive 1000 PCB coils? If so, could you arrange the coils in a matrix and reduce the number of FETs to 64?

[mikeselectricstuff]

The layers are mainly for PCB coils, which need to have fine-grained control, hence the mosfets. By using 2-channel SOT23-6L fets I am saving on both assembly part count and cost. I have the enable circuit down to ~$0.05 in BOM, there are just 350 copies of it...

Are you using 1000 FETs to drive 1000 PCB coils? If so, could you arrange the coils in a matrix and reduce the number of FETs to 64?

Even a moderate degree of muxing could cut it down a lot, both the MOSFETs and whatever I/O is driving them.
You may need more processing/IO speed/FPGA etc. but any additional cost for that will probably more than pay for itself.
For the manufacturing side it would definitely be worth spending a bit on a few hours' time with someone who has experience manufacturing similar scales of  product in that volume.

[mariush]

Think if it would be more cost effective to put your mosfets into double sided loads like DIMM sticks or M.2/mnipcie style cards. You could have them made on big panels and then use v-scoring to cut them to suitable sizes. These could be cheaper 2-4 layer circuit board instead of 8 layer boards, though I suspect your 8 layer board will be big due to the coils.

Also think if it would be possible to use ICs like ULN2003F12 (4 mosfets in 3.1mm x 3.1mm , but with common source, and built in gate resistors and esd protection etc) : https://www.digikey.com/en/products/detail/diodes-incorporated/ULN2003F12FN-7/5054226

Even at Digikey, it's 17 cents if you get 1000, so 4 cents per mosfet

The 7 channel ULN2003V12 is around 5.1mm x 6.4mm for the TSSOP package at 18.5 cents @ 1000 (so 2.5 cents per mosfet) : https://www.digikey.com/en/products/detail/diodes-incorporated/ULN2003V12T16-13/5054227


But maybe you can reduce from 8 layers to 4 or 6 and get cheaper board even with the extra headers?

[Smokey]

If you need 1000 MOSFETS and an 8 layer PCB for  "consumer game" product I suspect you might be doing something wrong.

That is the other big question, are you sure you have designed this with manufacturing, cost and simplicity in mind? Because as Mike points out, it doesn't sound like it.

I have spent quite a while optimizing for cost/simplicity already. The layers are mainly for PCB coils, which need to have fine-grained control, hence the mosfets. By using 2-channel SOT23-6L fets I am saving on both assembly part count and cost. I have the enable circuit down to ~$0.05 in BOM, there are just 350 copies of it...

It's a pretty unique design and I can't wait to announce it sometime next month.

PCB coils... Fet drivers... I'm going to take a guess here and put my money on a ferrofluid clock, or display, something like this:
https://www.thisiswhyimbroke.com/ferrofluid-clock/

32x32 grid gives you 1024 coils..

[jmelson]

If you need 1000 MOSFETS and an 8 layer PCB for  "consumer game" product I suspect you might be doing something wrong.

Yes, really!  8 layers, 450 x 420 mm!  I've never seen anything like that in consumer stuff!  You are going to be selling this gadget for $1000++, I hope!
Jon

[Infraviolet]

"8 layers, 450 x 420 mm!  I've never seen anything like that in consumer stuff!"
Isn't that about typical for the motherboards in some laptops? Though the manufacturer is likely using one motherboard design fora whole series of laptops with only minor changes between models, or just the soldering on or lack of particular components, so they get a lot of economies of scale.

That said, even then they're well below having 2000 components on such a board, much of the large area being to give more space for thermal considerations and to let the board run right up to the laptop's edges so all the connectors can be on one board and not need separate daughter boards wired to it.

[mariush]

A mATX motherboard is  9.6" by 9.6"  - 244 mm by 244 mm 

450 x 420 is nearly 4 mATX motherboards in a 2x2 arrangement... it's huge.