Alexander wrote "reducing load per one data bus pin". I guess he means Chip-Select lines.
Each memory rank (32 or 64 Bit wide depending on your setting) needs to be connected to a Chip-Select line.
On a 64 Bit bus you can either connect 8 pcs SDRAM components of a x8 config, or 4 pcs of x16, or 2 pcs of x32 to each chip-select-line.
Per my knowledge the capacitive load needs to be looked at "per each chip-select". The more components you use in parallel on the same chip select, the higher the load. This being said, you should be better off by using components in x32 width on as many chip-select lines as you have available.
If you want to maximize the amount of memory your system can support, you should use memory-components with smallest possible width per SDRAM.
From your other posting I see you plan to use 512Mb SDRAM in x16, IS45S16320F-7BLA2 in BGA54 package, Automotive Grade 2 (-40 to 105°C).
If your application requires high reliability and AEC-Q100 automotive grade, take a look at the ECC SDRAM series from Intelligent Memory. They have 512 Megabit capacity SDRAM components in x8, x16 and x32, with temperature ranges Industrial (-40 to 85°C), High (-40 to 105°C) and Extreme (-40 to 125°C).
The interesting thing about the IM ECC SDRAM parts is not only that they are available in AEC-Q100 and high temperature ranges, but they have a "built-in ECC error-correction engine" which detects and corrects bit-flips automatically. The parts still are fully compatible to any other SDRAM. There is no software or hardware adaption required. The error-correction is integrated into the components and executed without requiring any activity by the CPU/controller.
Honestly speaking, most normal automotive DRAMs are just "tested to work at a certain temperature". The operation at high temperatures is achieved simply by requiring the CPU/controller to double or quadruple the refresh-rate, because the data-retention-time of DRAMs is heavily reduced at higher temperatures.
Well, yes, with shorter refresh periods you can work against the reduced data-retention at high temperature. But what most DRAM manufacturers forget is that under high operating temperatures the DRAM memory cells (capacitors) degradate much faster, which - after some time of use - can cause an exponentially increased rate of transient single-event-upsets / bit flips. These bit-flips are not "defects", but "effects". An ECC error-correction is the perfect solution to achieve a stable 24/7 operation for many years. If your design supports ECC by the CPU, you better use that. Otherwise you have the option to use IM ECC DRAM components with built-in error-correction.
To make a simple example from real life:
Laptops or PCs need to be rebooted frequently. Is it really software fails?
Servers run the same Windows software and are under much more stress, but they work stable for years and never need a reboot. Reason: Servers use ECC error-correction!
On the Intelligent Memory website I currently see the 512Mb SDRAM x8 and x16 only in TSOP, but just yesterday I talked to them and I know for sure they are now releasing the same also in BGA54 package! I guess it will be on their website soon.
The 512Mb SDRAM in x32 comes in TSOP86 or in BGA90 and is shown on the website
Part numbers
512Mb SDRAM x8 IME5108SDBETG for TSOP54 IME5108SDBETG for BGA54 (BGA should be on website soon)
512Mb SDRAM x16 IME5116SDBETG for TSOP54 IME5116SDBETG for BGA54 (BGA should be on website soon)
512Mb SDRAM x32 IME5132SDBETG for TSOP86 IME5132SDBEBG for BGA90 (BGA should be on website soon)
Regards,
Thorsten