What type of technology are registers based on?

[chrstrbrts]

Hello,

Are registers based on SRAM, DRAM, or some other type of semi-conductor technology?

Thanks.

 

[sefsefsefsef]

SRAM. SRAM is very, very fast, and that's the only consideration when building registers.

 

[chrstrbrts]

SRAM. SRAM is very, very fast, and that's the only consideration when building registers.

Thanks.

 

[videogames101]

SRAM. SRAM is very, very fast, and that's the only consideration when building registers.

If he means a register file in a high-speed CPU core there's a good chance it's made of flops not SRAM.

 

[GAO]

Lol. What's a flop?

 

[videogames101]

Lol. What's a flop?

Flip Flops are two-state devices which can store a single bit of information.

(Faster than SRAM, less dense though)

 

[Wall Street]

SRAM is made of flip-flops - and registers are based on that tech too. You wouldn't say that registers are made out of SRAM because the acronym SRAM refers to random access memory and registers do not fill this role.

 

[videogames101]

SRAM is made of flip-flops - and registers are based on that tech too. You wouldn't say that registers are made out of SRAM because the acronym SRAM refers to random access memory and registers do not fill this role.

No.

This is a SRAM bitcell:

This is a (kind of) flip-flop:

SRAM is most certainly not made of flip-flops.

 

[Schmide]

SRAM is based on flip flops.

The above is a CMOS cell SRAM. It's just laid out as transistors rather than logic gates. A nand gate is just two transistors plus voltage control resistors.

 

[videogames101]

SRAM is based on flip flops.

The above is a CMOS cell SRAM. It's just laid out as transistors rather than logic gates. A nand gate is just two transistors plus voltage control resistors.

I assure you, a back to back inverter with drive lines (the SRAM cell) is NOT a flip-flop. I could draw both in transistors if you wanted, but maybe we're disagreeing on terminology? When anyone (at least in industry that I've met) says "SRAM" they mean a bunch of SRAM cells in an array, where each cell is a latch with some write and read transistors, not a flop-flop. When you're talking about the size of storage you need for a function on a given chip, it's flops OR SRAM. For a register file you could use either, but flops are getting more common I think.

Look up 6T/8T SRAM cells. No one refers to that as a flip-flop. =/

 

[TuxDave]

SRAM is based on flip flops.

The above is a CMOS cell SRAM. It's just laid out as transistors rather than logic gates. A nand gate is just two transistors plus voltage control resistors.

A standard design CMOS 2-input NAND gate is 4 transistors only.

And at least in my workplace, we view SRAMs and flip flops as two completely different things.

 

[Schmide]

An inverter is basically a NAND with both lines coupled. When you put two in series it becomes an AND.

AND - 2 NAND gates in series
OR - 2 NAND gates into a NAND
INV - NAND

I'm not a pro at this and this is from college 30 years ago.

The way I see it a sram cell is not a full latch it's kind of an optimized version of a latch. Replacing some of the resistors with transistors and forgoing the gates needed for proper assignment from a bus.

I like the way this guy put it all together.

http://electronics.stackexchange.com/questions/22130/various-memories/22135#22135

 

[Thala]

In our workspace we roughly distinguish (combinatorial) logic, registers (which could be either flip-flops or latches), clock-gates and SRAM.
The term registers is typically used in broader scope than just the CPU register file and refers to any kind of pipeline registers.

And while you can argue that both flip-flops and SRAM cells have 2 stable states driven by a feed-back, they are still completely different things. SRAM cells have a writeline (WL) input to enable either read or write using the same bitline (BL) as either input or output port, flip-flops have 2 separate ports for input and output and (typically) enable/clock input.

 

[sefsefsefsef]

I think people are getting confused by the terms "register" and "register file" in this topic. I thought it was pretty clear that the OP was referring to register files, which are absolutely made out of SRAM. The average computer enthusiast will maybe have some knowledge of the existence of the architected (and renamed) register file, but they probably won't know/care about the existence of other registers in the system. Register files are randomly accessible memory made out of flip flops, i.e., they are SRAMs.

Am I missing anything important there?

 

[KingFatty]

Yeah what registers are we talking about here? I assumed it was the CPU registers that you could use to store information, such as in assembly when you use a command such as: move eax, ebx

 

[Nothingness]

I think people are getting confused by the terms "register" and "register file" in this topic. I thought it was pretty clear that the OP was referring to register files, which are absolutely made out of SRAM. The average computer enthusiast will maybe have some knowledge of the existence of the architected (and renamed) register file, but they probably won't know/care about the existence of other registers in the system. Register files are randomly accessible memory made out of flip flops, i.e., they are SRAMs.

Am I missing anything important there?

Yes, reg files are not using SRAM for many reasons (latency, multiple accesses,,etc.). See this for instance: http://electronics.stackexchange.com/questions/72418/is-the-registry-file-made-from-sram

No design I have worked on in the past 15 years, from embedded to OoO engines, uses SRAM for reg files.

 

[Abwx]

The two circuits above are basicaly the same with the difference that :

SRAM store the value using mosfets input capacitance, wich mandate a periodic refresh..

Flip flop has an internal positive feedback loop, using two serialized inverters, to store the value, this has the advantage that a read is not destructive and refresh is not necessary, also the feedback loop increase the speed at wich the device stored state can be changed.

Edit : in the pic below the inverters are each no more than two complementary transistors.

 

[TuxDave]

SRAM store the value using mosfets input capacitance, wich mandate a periodic refresh..

I think you're thinking of DRAM. SRAM also has a positive feedback loop to store values.

 

[videogames101]

I think you're thinking of DRAM. SRAM also has a positive feedback loop to store values.

Confirming this.

Just to reiterate, although flip-flops are certainly "static" devices with inverter<->inverter positive feedback loops, they are not called SRAM. In industry SRAM refers very specifically to 6T/8T type bitcells with write and bit lines and auxillary circuitry like sense amps and sometimes fancy under/over supply voltage stuff to improve writability/stability of the bitcells. SRAM bitcells also have an inverter<->inverter feedback loop to store their values.

Flip-flops are edge sensitive (they have two feedback loops back to back), whereas SRAM cells are level sensitive with only one loop (looks like a latch). Also, flip-flops don't require external analogy circuits to function.

Today, registers and "register files" most often refer to flip-flops, and SRAM is used for larger memory arrays like an L1 or L2 cache where the extra analogy circuitry becomes a small part of the overall area compared to the bitcells in the array.

Hopefully the OP clarifies his question.