How much stuff are they cramming in there, really?

October 13, 2012 6:49 PM

Why do 32GB MicroSD cards cost more than 16GB cards, which cost more than 8GB cards, etc?

Is there something about the manufacturing process that makes producing them more expensive? Is there something about the raw materials involved that justifies the cost? Or is it a case where the people accept the logic that 32GB cards should cost more than 16GB cards and the manufacturers, understandably, are rolling with it?

posted by Golfhaus to Technology (16 answers total)

Well, there's more materials most likely, but that's not going to be your main reason. It's very, very likely that the difference in materials between a 32GB card and an 8GB card is pennies.

The reason for the size of the price differential is marketing. There's a certain subset of people who self-identify as wealthy, and buying an 8GB at $x isn't really palatable. Buying a 32GB at $z conforms to their mental image of themselves, so that's what they'll buy, even if they'll only ever use 4GB.

Conversely, there's people who are price-sensitive that you can still direct to the 16GB, because while the 32GB at $z seems WAY overpriced, the 16GB at $y seems like a steal.

In essence, it boils down to selling people on the perception of value, and for a great many people, buying a 32GB card at nearly the 8GB price would decrease the perceived value and thus satisfaction with the purchase.
posted by Imperfect at 7:05 PM on October 13, 2012 [3 favorites]

Also logically, the real question is why do 2GB or whatever the lowest one is, cost what they do? Because if we assume that their cost is (almost) justified by materials, then it makes sense that a double size one costs nearly twice as much and so on up the line. I suspect their cost is not justified by materials, and therefore Imperfect is right.
posted by lollusc at 7:07 PM on October 13, 2012 [1 favorite]

In a very simplistic sense, assuming the same process technology is used to make both a 16GB flash memory chip and a 32GB flash memory chip, the 32GB chip requires about twice as many transistors as a 16GB, hence about twice as much space on silicon wafer. So, in theory, a semiconductor fabrication plant that can produce a certain number of silicon wafers per month can produce twice as many 16GB flash memory dies as 32GB dies. The reason 32GB cards aren't twice the cost of 16GB cards is because lots of costs are associated with the number of devices produced, as opposed to their capacity, such as packaging the die, testing, SD card fabrication, consumer packing, etc.
posted by RichardP at 7:08 PM on October 13, 2012 [6 favorites]

It's the same reason people will upgrade to an XL drink for 40 cents more -- you're paying less for the extra amount than if you were to just get 2 of the smaller amount. Go, marketing.
posted by DoubleLune at 7:08 PM on October 13, 2012 [1 favorite]

Yes there is a function of the manufacturing process that goes into the cost of bit density. Each bit of that memory in a flash chip is a single floating gate transistor, of which there are huge shitloads on each chip. The chips are made with those transistors in large squares, so the area of the chip is directly related to the size. The size is fixed in each generation of semiconductor manufacturing, which occurs on large silicon wafers, and is done in cleanrooms to prevent contamination on the chips. The size of the transistors is small enough that small particles can destroy the viability of any single chip on that large wafer. To reduce that risk the chips are usually 2 or 4 GB in size. Thus you get 2 or 4 or 8 die in a package, sometimes a SD card, sometimes a SSD.

Many times, the size constraints for small package applications, like microSD cards, mean that the most expensive and smallest production techniques have to be used to make the die to be able to fit enough in the package. Most of the cost of semiconductor manufacturing is in the initial capital outlay, so thinking in terms of the materials used in production is not as relevant as thinking in terms of how much smaller they had to reduce the size from the older generations to make it work.
posted by apathy0o0 at 7:10 PM on October 13, 2012 [5 favorites]

Please read this. Electronic components like flash memory chips require incredibly extensive and precise manufacturing facilities. Breakthroughs are made regularly in increasing storage density, but manufacturing processes for new and untested technologies are always more expensive and less refined than those for old, established technology. It has nothing to do with marketing. It has everything to do with industrial manufacturing and competition and innovation among suppliers.
posted by tehloki at 7:15 PM on October 13, 2012 [8 favorites]

It is overly simplistic to look at the unit cost on a raw materials + labor basis. On a day to day snapshot basis, it might only cost the factory a nickel to make each unit. But you have count what it cost to build the factory and invent the technology to make the thing. In industries like this, where it takes a shit-ton of R&D to get the product ready for production, those costs are amortized over the production run.

Also, it is more expensive to fabricate more complicated silicon wafers. Every transistor is another point of failure. It's a one bad apple spoils the bunch problem. The bigger each bunch is, the more likely there is to be a bad apple in one.

You have to count the wasted cost of building the bad ones toward the cost of building the good ones.

Another facet of the issue is that the newest processes make the smaller capacity ones cheaper to build. A 32gb wafer with a defect at the 24gb spot can be "neutered" and sold as a 16gb SD card with no problems. And every 32gb card they sell is a 16gb card they didn't, eroding the price for the 16gb cards.

Marketing plays a part in the process, but not in the simple, cynical way some might think. Marketing just figures out what price the market will bear. There still has to be A market for the product and there will be competitors.
posted by gjc at 7:41 PM on October 13, 2012 [5 favorites]

So, in theory, a semiconductor fabrication plant that can produce a certain number of silicon wafers per month can produce twice as many 16GB flash memory dies as 32GB dies.

It's even worse than that. There are a certain number of process flaws per wafer, and each process flaw destroys one chip. If you get 100 16G chips per wafer and lose 5 to process flaws, then your yield is 95%. If you get 50 32G chips per wafer and lose 5, your yield is 90%. (Those numbers are not representative; used only for explanatory purposes.)

So the yield of 32G chips is less than 1/2 the yield of 16G chips.
posted by Chocolate Pickle at 8:02 PM on October 13, 2012 [3 favorites]

The more parts an item has the more likely it is to have manufacturing defects.
A memory card with 16 billion microscopic parts is going to have more defects than one with merely 8 billion microscopic parts. Therefore they will have to either make additional units to cover the defects, or take significantly more care when making them to keep defects low. Either choice makes them more expensive.
posted by Ookseer at 8:06 PM on October 13, 2012

What RichardP said. If the chips are made using the same process, then the lower-capacity chips will be physically smaller, which means they'll take less space on the wafer, which means more of them per wafer, which means that their share of the (rather large) cost of running one wafer through the fab is smaller.

if the chips are made using different processes, the lower-capacity chip is probably made using an older, less-dense process whose fab is older (therefore, closer to being paid off), in less demand, and possibly has a higher yield since it's been around longer.

Some fraction of 16GB chips might be 32GB chips with a manufacturing defect, which have had the defective bank turned disabled and are sold as 16GB chips— I don't know how common that is; as I understand it, large memory chips usually have a small amount of excess capacity which is used to replace defective sections, but I wouldn't expect that to be a whole half of the chip.

There are also a bunch of costs independent of size, like packaging and testing and shipping and retailing, which probably provide a price floor for the lowest-capacity cards.
posted by hattifattener at 8:09 PM on October 13, 2012

There are a certain number of process flaws per wafer, and each process flaw destroys one chip. If you get 100 16G chips per wafer and lose 5 to process flaws, then your yield is 95%. If you get 50 32G chips per wafer and lose 5, your yield is 90%.

This may be the case for certain chips like CPUs or DRAM but not for Flash memory chips. Flash memory chips have spare blocks which are used to replace defective blocks. The internal block mapping bypasses these just like bad sectors on a disk drive. Whether you have one bad block on a small Flash chip or two bad blocks on a larger Flash chip, both are serviceable chips because the percentage of defective blocks is the same. In other words, for Flash memory, the granularity of defects is by block size, not by chip size.
posted by JackFlash at 11:45 AM on October 14, 2012

One aspect yet unmentioned is that storage (in general, memory cards, disk drives, RAM) is typically a commodity. There are numerous manufacturers competing to offer the highest capacity at the lowest cost, and it's not really possible for one to affect the market unless they collude. There are a few other exceptions where price doesn't quite follow capacity, such as Apple's pricing for flash storage in iPod/iPhone/iPad, but the market is generally extremely competitive.
posted by wnissen at 8:13 PM on October 14, 2012

For both software and silicon chips, the $$ required to get to your first sellable product is incredibly high, while each additional unit is rather cheap in comparison. You seem to think that the fair price should be some small multiplier of the basic labor and materials the cost for each incremental unit, but obviously the cost of each incremental unit has to cover the cost of producing the first units.

Semiconductor manufacturers spend a lot of money (billions of dollars) on latest generation fabrication plant. They also spend tens of millions, if not more, designing the chips that get built in those plants, and, I expect, millions more putting a new design into production in a given plant. Figuring out how to pay back those costs and, hopefully, earn a profit is not a simple problem.

However, even at the simplest level every 32GB flash chip they sell is a lost opportunity to sell 2 16GB chips, or 4 8GB chips, isn't it evident that the pricing should increase proportionally? Or are you actually asking why it sometimes seems that it costs more than 4x as much to get a 32GB flash card vs a 8 GB flash card?

These things happen for a variety of reasons, but generally it comes down to people being willing to pay a premium for the convenience or longer useful life of larger/faster chips, and companies taking advantage of that to optimize their profit margins and return on investment.
posted by Good Brain at 8:51 PM on October 14, 2012

Chocolate Pickle: So, in theory, a semiconductor fabrication plant that can produce a certain number of silicon wafers per month can produce twice as many 16GB flash memory dies as 32GB dies.

It's even worse than that. There are a certain number of process flaws per wafer, and each process flaw destroys one chip. If you get 100 16G chips per wafer and lose 5 to process flaws, then your yield is 95%. If you get 50 32G chips per wafer and lose 5, your yield is 90%. (Those numbers are not representative; used only for explanatory purposes.)

So the yield of 32G chips is less than 1/2 the yield of 16G chips.

This is the correct answer.

Source: 4 yrs working at ASML, the largest producer of photolithography chip equipment - they make over half the machines in the world that make computer chips.
posted by IAmBroom at 8:55 AM on October 15, 2012

IAmBroom, as explained above, this factor applies to chips like CPUs or DRAMs but less so to Flash memory which is designed with spare blocks and expected to have a few bad blocks, just like a regular disk drive. A defect on a Flash chip generally only affects a single block. One bad block on a small die vs two bad blocks on a die twice the size are equivalent and does not imply a lower yield. For Flash memory, the granularity of defects is not chip size but block size and is the same in both cases.

You can get a flavor for the details here.

It is possible to have a defect that would disrupt the control circuitry for the entire die, but since Flash chips consist predominantly of reiterated storage cells and only a tiny percentage of global control circuitry, global defects that invalidate an entire die have only a miniscule effect on overall yield.
posted by JackFlash at 9:51 AM on October 15, 2012 [2 favorites]

Dammit! I stand corrected, JackFlash. My concentration was on logic devices, not memory, and you are probably correct.
posted by IAmBroom at 10:12 AM on October 15, 2012