Second-gen Phison SSD controllers hit the market with 14.9 GB/s speeds — TeamGroup Z54E spearheads a new army of PCIe 5.0 drives

(Image credit: TeamGroup)

TeamGroup has announced its new T-Force Z54E, a PCIe 5.0 SSD designed to trade blows with the best SSDs on the market. Notably, the Z54E isn't just another PCIe 5.0 drive but the first retail SSD to launch with Phison's high-performance E28 SSD controller, setting it apart from the competition.

The E28 controller signifies a substantial advancement over its predecessor, the E26 controller. Utilizing TSMC's 6nm process technology, the E28 delivers enhanced performance and superior power efficiency. Furthermore, the E28 serves as Phison's response to Silicon Motion's SM2508, which has been expanding its presence within the PCIe 5.0 market segment. With the introduction of the E28, Phison endeavors to reestablish its position as the leading manufacturer of PCIe 5.0 controllers.

The Z54E will be one of many consumer PCIe 5.0 M.2 2280 drives employing the Phison E28 controller. TeamGroup does not specify the particular type of flash memory used in the Z54E, only indicating that it is constructed with "high-density 3D TLC NAND flash." If an inference must be made, it is likely to utilize a 218-layer NAND, suggesting that TeamGroup may potentially be sourcing the flash from Sandisk or Kioxia.

TeamGroup T-Force Z54E Specifications

Swipe to scroll horizontally

Capacity	Sequential Read (MB/s)	Sequential Write (MB/s)	DRAM Cache (GB)	Endurance (TBW)
4TB	14,900	14,000	4	2,400
2TB	14,900	14,000	2	1,200
1TB	14,900	13,700	1	600

The Z54E delivers sequential read and write speeds up to 14,900 MB/s and 14,000 MB/s, respectively, in its 4TB and 2TB configurations. The 1TB model, representing the lowest capacity SKU, operates with a sequential write speed approximately 300 MB/s lower. TeamGroup has not disclosed the random performance metrics for the Z54E. The drive is equipped with a DRAM cache; 1GB of cache per 1TB of capacity. Additionally, the company has not released the random performance figures for the Z54E.

Endurance performance on the Z54E is not too bad. The 1TB model is rated for 600 TBW, a figure that is subsequently doubled as capacity increases. The endurance levels are comparable to those of competing drives, such as the Sandisk WD Black SN8100 and Samsung 9100 Pro.

Image 1 of 4

The Z54E will be equipped with TeamGroup's patented ultra-thin graphene heatsink to enhance heat dissipation. It appears that the company is not providing a version incorporating its T-Force Dark Flow active SSD cooler on this occasion. Considering that the E28 draws a maximum of approximately 7W, the Z54E is likely adequately cooled without the need for a chunky active cooling solution.

TeamGroup supports the Z54E with a limited five-year warranty. The company did not disclose the pricing or availability; however, we have contacted them for further information. We will provide updates to this report once the company responds.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Zhiye Liu is a news editor, memory reviewer, and SSD tester at Tom’s Hardware. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

7 Comments Comment from the forums

Stomx

"Utilizing TSMC's 6nm process technology..."
Why not 5, 4, 3, 2, 1.x if it is so critical to overheating?
Ahhh, ooops, sorry, I forgot ... planned obsolescence. Worst of the worst thing for the manufacturers is to produce ideal eternal device
Reply
thestryker

Stomx said:
"Utilizing TSMC's 6nm process technology..."
Why not 5, 4, 3, 2, 1.x if it is so critical to overheating?
Ahhh, ooops, sorry, I forgot ... planned obsolescence. Worst of the worst thing for the manufacturers is to produce ideal eternal device
It has nothing to do with "planned obsolescence" and everything to do with low margin parts on high cost nodes. I imagine if AMD/nvidia had moved on from using a TSMC N5 based node for their most recent generation one of those may have been an option as prices would have dropped. That didn't happen though so N6 is the best node performance wise that isn't carrying a big price premium.
Reply
Amdlova

It's better them in old process... Cheaper and last more time...
Reply
Stomx

thestryker said:
It has nothing to do with "planned obsolescence" and everything to do with low margin parts on high cost nodes. I imagine if AMD/nvidia had moved on from using a TSMC N5 based node for their most recent generation one of those may have been an option as prices would have dropped. That didn't happen though so N6 is the best node performance wise that isn't carrying a big price premium.
Interesting how do you estimate cost and their margin both for NAND and for example AMD EPYC
Reply
thestryker

Stomx said:
Interesting how do you estimate cost and their margin both for NAND and for example AMD EPYC
I don't really understand your question here.

NAND isn't made on TSMC nodes as Samsung, SK Hynix and Micron all have their own fabs for memory IC as it requires a lot of specialization.

As for the controller itself they're not particularly small. According to TPU the NAND size is ~60mm^2 and in the teardown pictures of the same drive the controller is bigger than the NAND. That means the E28 on N6 the die most likely has a least the same area as a Zen 3 CCD. One just has to consider SSD pricing compared to CPU pricing to see where the margin problem starts to come in since SSDs also have NAND and DRAM as well.
Reply
bit_user

thestryker said:
As for the controller itself they're not particularly small. According to TPU the NAND size is ~60mm^2 and in the teardown pictures of the same drive the controller is bigger than the NAND.
The package is bigger than the NAND, but I think we don't know what was the limiting factor. Could be I/Os? If it's really the die size, I do wonder why. I guess the PCIe 5.0 controller will eat a chunk, but the cores are listed as being only "ARM 32-bit Cortex-R5", which should be much smaller than ARM's 64-bit E-cores. Might it harbor a big chunk of SRAM, in support of DRAM-less applications? It seems to me that the amount of SRAM you need for that should be proportional to the amount of NAND the drive has. So, a newer controller that's planned to support higher-capacity SSDs probably needs to have more SRAM.

Another factor favoring a larger package is that it lets them use a bigger heatspreader. I think the natural limit on controller package size is just leaving enough room on a 2280 drive for a DRAM package and two NAND packages.

BTW, don't forget that the NAND is usually die-stacked, in addition to having multiple layers per die. That makes comparison of package sizes somewhat deceiving.
Reply
thestryker

bit_user said:
The package is bigger than the NAND, but I think we don't know what was the limiting factor. Could be I/Os? If it's really the die size, I do wonder why. I guess the PCIe 5.0 controller will eat a chunk, but the cores are listed as being only "ARM 32-bit Cortex-R5", which should be much smaller than ARM's 64-bit E-cores.
I'd love to see a die shot of one of these controllers because I know the CPU part should be relatively tiny. In one of the videos I watched with Allyn Malventano (I think it was after he went to Phison) he definitely mentioned that all the engineers really wanted more advanced nodes for the controllers but those always came later. I think that poor timing is why Silicon Motion was able to get a big jump on the second generation of PCIe 5.0 M.2 SSDs (unless Phison just sold the first batches to enterprise).

bit_user said:
Might it harbor a big chunk of SRAM, in support of DRAM-less applications?
I don't believe the E28 supports DRAM-less operation so it shouldn't be that. Once Phison puts up their datasheet it should be possible to get a better idea as to what all there is.

Silicon Motion's SM2508 is a quad core Cortex-R8 and uses the same 576 ball count package (different physical size) as Phison's existing E26 (Dual Cortex-R5). The reason I mention that is the E31T (Phison's DRAM-less PCIe 5.0 controller) halves the maximum NAND channels, uses a single core architecture, and uses a 228 ball package. Perhaps there's enterprise features on the controllers which aren't exposed on client drives which are also increasing the size.
Reply

Show more comments