Is it cheaper to manufacture a embedded chip or a chip with removable parts?

[Flux]

Is it cheaper to manufacture a single board SoC or a chip with removable parts?

Let's say if you build a computer with chips that are all on the same board and not removable
from the computer. You know, on-die eRAM(22nm),SoC/PiP(22nm) processor,cache and other support ICs,integrated southbridge/northbridge, and a sound small dsp.

Will that be cheaper to build in mass compared to a standard removable PC components?

 

[3dcgi]

It's going to depend a lot on how big the piece of silicon is for the SoC. There's overhead to individual chips as they need pins and if there's not enough logic the pins or more likely pads dictate the size of the chips. So it's very possible for the SoC to be cheaper and more power efficient to boot.

If the SoC is large the advantage of being a single piece of silicon is diminished or non-existent. There are multiple variables affecting what's best.

 

[Flux]

So DIMM ram is cheaper than ram that is wired on the mainboard?

If the components are embedded on the chipset is it cheaper compared to a computer that has parts that are removable?

 

[Silent_Buddha]

So DIMM ram is cheaper than ram that is wired on the mainboard?

If the components are embedded on the chipset is it cheaper compared to a computer that has parts that are removable?

From a pure cost perspective, no. It's cheaper to have the ram, CPU, BIOS, whatever, directly soldered to the board.

From an implementation perspective, however, that isn't necessarily the case.

For an OEM the flexibility when offering a product line means that over the lifetime of those products it can be cheaper to go with socketed designs even it costs more per part.

For example, with RAM. You don't have to attempt to predict demand for 2 GB, 4 GB, 8, GB, 16 GB, etc. systems. Hence you don't have to worry about having say, excess stock of mainboards with 2 GB of RAM. Situations like that can end up costing you more money long term than saving a bit on the cost of a DIMM slot.

You could always limit your stock significantly and/or only manufacture the parts when required. But then you lose things like economies of scale as well as potentially having shortages of a specific configuration at which point a potential customer may go to a competitor instead of waiting.

Speaking of economies of scale. That's another benefit of socketed systems. You can have, for example, one socketed mainboard that services multiple product lines allowing you to mass produce those mainboards in significantly more quantities and with associated cost savings over multiple specialized mainboards with everything soldered on.

Then again for something like embedded systems where they are generally made to order or only offered in predefined configurations, the flexibility offered in system integration by having interchangeable parts isn't necessarily required and would just add extra cost for no cost benefit.

Regards,
SB

 

[Flux]

Thank you for your response.

 

[tunafish]

Is it cheaper to manufacture a single board SoC or a chip with removable parts?

Buddha answered for the board perspective, I'll give the chip one. The fundamental driving principle of the silicon industry is that the cost of producing a chip of ~the same dimensions is the same, but transistors keep getting smaller. There is an ideal size for a chip where each transistor is at it's cheapest, dictated by many factors (reticle size, yields, etc), and what you can fit in this size doubles every new full node.

At first, just a single CPU core itself could use all the transistors we could fit on an ideal-size chip -- each new generation efficiently used more transistors for more effect. Then the industry more or less ran out of ideas, and now putting more transistors in the cpu just buys minor incremental gains per transistor added. At this point, instead of using the transistors for the core alone, it made sense to start integrating other things, like the memory controller. Then one core became two, or more, and the entire north bridge and the gpu migrated on die. Now on the newest low-power chips most of the south bridge is integrated too.

If Moore's law doesn't halt before it, or we don't get great new ideas about how to use more transistors, eventually it will make sense to bring the memory on die too.

 

[Flux]

Buddha answered for the board perspective, I'll give the chip one. The fundamental driving principle of the silicon industry is that the cost of producing a chip of ~the same dimensions is the same, but transistors keep getting smaller. There is an ideal size for a chip where each transistor is at it's cheapest, dictated by many factors (reticle size, yields, etc), and what you can fit in this size doubles every new full node.

At first, just a single CPU core itself could use all the transistors we could fit on an ideal-size chip -- each new generation efficiently used more transistors for more effect. Then the industry more or less ran out of ideas, and now putting more transistors in the cpu just buys minor incremental gains per transistor added. At this point, instead of using the transistors for the core alone, it made sense to start integrating other things, like the memory controller. Then one core became two, or more, and the entire north bridge and the gpu migrated on die. Now on the newest low-power chips most of the south bridge is integrated too.

If Moore's law doesn't halt before it, or we don't get great new ideas about how to use more transistors, eventually it will make sense to bring the memory on die too.

Oh ok.

Thank you for responding.