The most expensive SRAM in the world – 2.0

By Zvi Or-Bach, President and CEO of MonolithIC, and Ben Louie, Zeno Semiconductors

ISSCC 2014 illuminates the impeding problem – embedded SRAM scaling. The following two slides are taken from the Dinesh Maheshwari, CTO, Memory Products Division at Cypress Semiconductors, presentation. The first slide clearly illustrates that embedded SRAM scaling is broken. Instead of 4X density improvement for a large memory block for two nodes scaling the improvement range is only 1.6X for low performance to 1.1X at good performance.


Maheshwari’s following slide assesses the implications for 400G ASSP/ASIC. He concludes that it impractical to have the eSRAM integrated; which mean that Moores’ Law is actually broken for these type of applications.


Taking in account that 16/14nm silicon area is almost twice as expensive as that of 28nm, it implies that the embedded SRAM within 16/14nm ASSP/ASIC will become the most expensive SRAM in the world. Interestingly, the title we used in a blog in mid 2011  ”The most expensive SRAM in the world” was made in reference to the embedded SRAM in Intel’s processors.

The percent of the die area used for embedded SRAM is growing with scaling and already exceeds 50%. The following two charts from Semico, which were recently updated, illustrates this for an advanced SoC and the average SoC.



Following the previous ISSCC 2013 we wrote a blog -One Thing that ISSCC 2013 Highlighted to Us, which identified the problems associated with embedded SRAM scaling. This year’s ISSCC 2014 just added additional validation to the issue.

It interesting to look at the following chart which presents the prior belief that SRAM will keep scaling at 50% per node:


*from the book by: M. H. Abu-Rahma and M. Anis, Nanometer Variation-Tolerant SRAM,  © Springer Science+Business Media New York 2013

The above chart was about right all the way to 28nm, but scaling has broken since then. Already at 22/20 nm node the best bitcell size was about 0.09µm². But as TSMC reported at IEDM 2013 their bitcell for 16nm is 0.07µm². And now at ISSCC2014, Samsung presented similar results for 14nm FinFet as shown in the slide below.


And it seems that these metrics become far worse when comparing the size of high performance block RAM between technology nodes as been presented above by Maheshwari.

This does not even take into consideration die size impacts required for implementing a new technology such as FinFETs.  In their ISSCC 2014 paper, Samsung identified complications involved with the transition from Planar to finFET transistors including quantized width, strong PMOS and a lack of the body bias effect.  Samsung disclosed a die size impact of 0.87% in order to add a Disturbance Noise Reduction (DNR) scheme to deal with the large more stable High Performance (HP) bit-cell.  Samsung did not disclose the area impact of their proposed Negative Bitline write assist scheme used with their high density HD bit-cell but we can probably assume it is most likely significantly larger than the proposed scheme for the HP bit-cell.

It is not too surprising to see the following slide presented by Intel at ISSCC 2014.


So while Moore’s Law is about the optimal integration into one IC, Intel has chosen to forgo single IC integration and is instead turning to external memory to provide their needs.

We can see from the following slide from ISSCC 2014 their motivation for adding the off chip embedded eDRAM is based on their need for a higher bandwidth Memory.  Clearly SRAM scaling is not satisfying their requirements as they state “A high-density, high BW In Package Memory is needed.”


From what we have learned there are few issues that hold back the SRAM bitcell scaling:

A. Lithography, the continuous delays in EUV force the industry to keep pushing with double and quad processes. The following chart by ASML illustrates the limitation effects on SRAM scaling:


B. Increase random variations as transistors are getting smaller


*from Prof Nam Sung Kim presentation titled: SRAM Scaling Limit: Its Circuit & Architecture Solutions

C. Decreasing VCC

As seen in the above slide, and the slide presented by Samsung below, the trend is definitely clear.  The minimum Vdd require to operate SRAM (Vmin) is not scaling as fast as the Vdd in the rest of the logic on die.


D. Other issues such as increased Soft Error Rate with reduce transistor sizes, and increases in overall wire capacitance and resistance.

Bitcell scaling is getting harder, much harder, and even more so is the ability to scale large bocks of embedded SRAM. Adding to that the trend of growing the amount of embedded memory faster than the amount of logic cells and one could predict dark clouds for SoC scaling beyond 28nm. It seems that brute-force scaling is not practical anymore but two technology innovation breakthroughs could solve the SRAM memory scaling problem and provide a scalable high density memory if adopted soon by the industry.

The first innovation is the 1 Transistor SRAM (1T SRAM) developed by Zeno Semiconductor. This 1T SRAM, utilizes existing fab process and provides a 90% bitcell size reduction vs. conventional 6T SRAM and will keep scaling beyond 28nm.

The second is monolithic 3D, which enables a very effective heterogeneous integration scheme, allowing for the SRAM layer to be optimized for memory while the logic layer could be optimized for logic – Monolithic 3D eDRAM on Logic

The most expensive SRAM in the world doesn’t need to come to pass.


Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>


SEMI-GAS broadens gas mixing capabilities for highly corrosive gases
08/28/2014SEMI-GAS Systems, a provider of ultra-high purity gas delivery equipment, recently broadened the capabilities of its custom Xturion ...
Entegris announces GateKeeper GPS platform
07/15/2014Entegris, Inc., announced last week the launch of GateKeeper GPS, its next-generation of automated regeneration gas purification system (GPS) technology....
Bruker introduces Inspire nanoscale chemical mapping system
07/15/2014Bruker today announced the release of Inspire, the first integrated scanning probe microscopy (SPM) infrared system for 10-nanometer spatial...