Exynos 7420: First 14nm Silicon In A Smartphone

This generation more than any generation in recent memory has been a time of significant movement in the SoC space. We were aware of the Exynos 7420 well before it was announced in the Galaxy S6, but for the most part I expected to see Snapdragon 810 in at least a few variants of the Galaxy S6. It was a bit surprising to see Samsung drop Snapdragon SoCs completely this generation, and judging by the battery life of the Galaxy S6 it seems that Samsung had their reasons for doing this.

For those that are unfamiliar with the Exynos 7420, this SoC effectively represents the culmination of their efforts in semiconductor manufacturing and integrated circuit design. On the foundry side, Samsung is leveraging their vertical integration to make the first SoC on their 14nm LPE (Low Power Early) process, which seems to be solely for Systems LSI until they can no longer use all production capacity.

We previously mentioned that Samsung’s 14nm process in general will lack any significant die shrink due to almost unchanged metal interconnect pitch, but this assumption was in comparison to their 20nm LPM process from which the 14nm LPE process borrows its BEOL (back end of line) from. Opposite to what we thought, the Exynos 5433 was manufacturered on a 20LPE process which makes use of a quite larger metal layer. The result is that one can see a significant die shrink for the 7420 as it is, according to Chipworks, only 78mm² and a 44% reduction over the Exynos 5433's 113mm². This is considerable even when factoring in that the new SoC had two added GPU shader cores. Beyond the swap from a LPDDR3 memory controller to a LPDDR4 capable one, the only other at first noticeable major functional overhaul on the SoC seems to be that the dedicated HEVC decoder block has been removed and HEVC encoding and decoding capability has been merged into Samsung's MFC (Multi-Function Codec) media hardware acceleration block.


Galaxy S6 PCB with SoC and modem in view (Source: Chipworks)

The move from a planar to FinFET process is crucial. Although this is covered in more detail in previous articles, the short explanation is that planar processes suffer from increasing power leakage at smaller process nodes as the bulk of the silicon becomes relatively more massive than the gate that controls the flow of current. This causes decreased power efficiency as the power source of the transistor starts to act as a gate itself. FinFET solves this problem by attempting to isolate the transistor from the bulk of the silicon wafer, wrapping the gate around the channel of the transistor to ensure that it retains strong control over the flow of current compared to a planar transistor design.

The effective voltage drop allowed by the process can be substantial. We can have a look at some voltage excerpts of common frequencies available on both the Exynos 5433 and 7420:

Exynos 5433 vs Exynos 7420 Supply Voltages
  Exynos 5433 Exynos 7420 Difference
A57 1.9GHz (ASV2) 1287.50mV 1056.25mV -234.25mV
A57 1.9GHz (ASV9) 1200.00mV 975.00mV -225.00mV
A57 1.9GHz (ASV15) 1125.00mV 912.50mV -212.50mV
A57 800MHz (ASV2) 950.00mV 768.75mV -181.25mV
A57 800MHz (ASV9) 900.00mV 687.50mV -224.50mV
A57 800MHz (ASV15) 900.00mV 625.00mV -275.00mV
A53 1.3GHz (ASV2) 1200.00mV 1037.50mV -162.50mV
A53 1.3GHz (ASV9) 1112.50mV 950.00mV -162.50mV
A53 1.3GHz (ASV15) 1062.50mV 900.00mV -162.50mV
A53 400MHz (ASV2) 862.00mV 743.75mV -118.25mV
A53 400MHz (ASV9) 787.50mV 656.25mV -131.25mV
A53 400MHz (ASV15) 750.00mV 606.25mV -143.75mV
GPU 700MHz (ASV2) 1125.00mV 881.25mV -243.75mV
GPU 700MHz (ASV9) 1050.00mV 800.00mV -250.00mV
GPU 700MHz (ASV15) 1012.50mV 750.00mV -262.50mV
GPU 266MHz (ASV2) 875.00mV 750.00mV -125.00mV
GPU 266MHz (ASV9) 800.00mV 668.75mV -131.25mV
GPU 266MHz (ASV15) 762.50mV 606.25mV -156.25mV

The ASV (Adaptive Scaling Voltage) numbers represent the different type of chip bins, a lower value representing a worse quality bin and a higher one a better quality one. Group 2 should be the lowest that is found in the wild, with group 15 representing the best possible bin and group 9 the median that should be found in most devices. As one can see in the table, we can achieve well up to -250mV voltage drop on some frequencies on the A57s and the GPU. As a reminder, power scales quadratically with voltage, so a drop from 1287.50mV to 1056.25mV as seen in the worst bin 1.9GHz A57 frequency should for example result in a considerable 33% drop in dynamic power. The Exynos 7420 uses this headroom to go slightly higher in clocks compared to the 5433 - but we expect the end power to still be quite lower than what we've seen on the Note 4.

On the design side, Systems LSI has also done a great deal to differentiate the Exynos 7420 from the 5433. Although the CPU architectures are shared, the A53 cluster is now clocked at 1.5 GHz instead of 1.3 GHz, and the A57 cluster at 2.1 GHz rather than 1.9 GHz. The memory controller is new and supports LPDDR4 running at 1555MHz. This means that the Galaxy S6 has almost double the theoretical memory bandwidth when compared to the Galaxy Note 4 Exynos variant, as we get a boost up to 24.88GB/s over the 5433's 13.20GB/s. We still need to test this to see how these claims translate to practical performance in a deep dive article in the future, as effective bandwidth and latency can often vary depending on vendor's memory settings and SoC's bus architecture. 

Outside of the memory controller, LSI has also updated the 7420 to use a more powerful Mali T760MP8 GPU. Although the Exynos 5433 had a Mali T760 GPU as well, it had two fewer shader cores which means that achieving a given level of performance requires higher clock speeds and higher voltages to overcome circuit delay. This new GPU is clocked a bit higher as well, at 772 MHz compared to the 700 MHz of the GPU in the Exynos 5433. We see the same two-stage maximum frequency scaling mechanism as discovered in our Note 4 Exynos review, with less ALU biased loads being limited to 700MHz as opposed to the 5433's 600MHz. There's also a suspicion that Samsung was ready to go higher to compete with other vendors though, as we can see evidence of an 852 MHz clock state that is unused. Unfortunately deeply testing this SoC isn’t possible at this time as doing so would require disassembling the phone.

Introduction and Design Battery Life and Charge Time
Comments Locked

306 Comments

View All Comments

  • JoshHo - Friday, April 17, 2015 - link

    Something similar to the APx582 would be necessary to test 3.5mm output accurately, and a good SPL meter would help with speaker quality although I'm not sure results for the latter would be comparable between editors.
  • IsthatyouBevis - Friday, April 17, 2015 - link

    Wow, that camera comparison is really poorly done. Trying to compare what you seem to think are the 2 best cameras (iPhone 6+ and S6) was extremely difficult. Why do you present the galleries in such a way as it is impossible to determine which picture to click to see which phone and why can't we see them clearly in the same page, instead of the click fest you force us to go through? Surely there is a better way to demonstrate this?
  • johnnohj - Friday, April 17, 2015 - link

    1. Were these photos taken with Auto-HDR enabled? If I am not mistaken, both the iPhone and S6 ship with Auto-HDR turned on by default. Those impressive low-light photos Samsung showed at the unveiling were taken with Auto-HDR enabled.

    2. Can immersive mode (full-screen) be disabled on Samsung apps like Calculator and S-Planner?

    3. Were there any problems with RAM management like in this video https://youtu.be/hUw9PUlFUF0?t=1m32s where the S6 keeps killing apps in the background? It could explain the poor battery life some people are experiencing as apps would have to be killed and reloaded all day.

    4. Can heads-up notifications be disabled?

    5. I wish there was a way to test standby battery usage. It is something android phones have always been poor at compared to iPhones. How about a test where the x most popular apps (like Facebook, Gmail, WhatsApp, Instagram, Twitter with the same accounts, sync enabled) are installed to multiple test devices on the same wifi network and the battery usage is measured after x hours?

    6. Why does Samsung still include the Quick Connect and S-Finder buttons in the notification slider? It's so annoying, and there is no way to disable them without root (except on a few US carrier versions).
  • Hrel - Friday, April 17, 2015 - link

    No sd card is pretty bad, but being Samsung is worse.
  • khujin - Friday, April 17, 2015 - link

    I just tested GS6 edge 64GB and Androbench4.0 result is 320MB/s(seq. read), 140MB/s(seq. write) with default setting. And another website GS6 performance review it the same as mine. please check the UFS test.
  • JoshHo - Friday, April 17, 2015 - link

    The default sequential test does 32MB sequential reads which is wildly unrealistic at best. As shown in the storage performance section we test using 256KB reads and writes for sequential performance.
  • heartinpiece - Saturday, April 18, 2015 - link

    Nice review Joshua! I don't see any Ipad Air 2 performances in the performance charts (although they are constantly referenced...) could you check on the graphs?
  • Peichen - Saturday, April 18, 2015 - link

    Seems like there are only 2 company making real flagship phones now: Apple and Samsung. To achieve this status, buying off-the-shelf parts are no longer good enough. You need to custom order parts and optimize the software specifically for the parts. From SoC, memory, screen to camera, everything is custom ordered to achieve the best quality. Off-the-shelf phones like M9 and Nexus 6 are relegated to good $500ish phone status.

    On the other hand, it is sad to see Samsung abandon all its believes and completely adapted Apple design and philosophy. Nano-SIM, downward speaker, simplified camera UI, fingerprint magnet back, home-button scanner and the abandonment of colorful UI and screen tone. Samsung users always love the punchy screen color. What are they going to say now the school is color corrected like iPhone display?
  • Peichen - Saturday, April 18, 2015 - link

    screen*
  • Impulses - Saturday, April 18, 2015 - link

    Maybe, I think LG or Sony could still step up to the plate... Sony's got a whole bunch of outside issues threatening the future of it's phone division tho, and probably their budget. Which is a shame because they made the only SMALL flagship phone last year in the Z3c.

    The bigger issue at play might be that a good chunk of the market is now complacent with their $500 or less also-rans, and another sizeable chunk is just buying outgoing models instead of the current hotness. Otherwise the SGS5 should've sold even better than the 4, and it had every right to based on hardware alone.

    I'm part of that segment tbh, huge Android fan, initially went thru three successive HTC phones year after year... Not really feeling that upgrade bug anymore, at all. I'd like better battery life than what my N5 offers, and/or water resistance, can live without either for now.

    My phone isn't my main camera so I'm an exception there but I don't think the average buyer even realizes what makes for a better camera or reads AT to find out.

Log in

Don't have an account? Sign up now