Sam Beal & Associates

Technology Marketing - Marketing Technology

Archive for the Technology Category

NAND Flash Ecosystem

NAND Flash is the dominant non-volatile storage technology in portable electronics (Smartphones, Tablets). Solid state drives (SSD) are rapidly displacing hard disk drives (HDD) in laptops and moving into enterprise storage in a variety of configurations – flash cache, hybrid and full flash storage arrays.

I created a an introductory slidedeck of the ecosystem circa mid 2014. You can view/download it from slideshare.

Why the Aereo Supreme Court Decision is ultimately good for consumers

You can read the full transcript of the SCOTUS decision here.

Here is my simple interpretation.

Over the air TV – which is strictly local – is free (unless you count your time spent watching Advertising). The local broadcasters make money by selling “local” ads. If a service can redirect the ATSC feed to a TCP/IP stream, then the value of local advertising is lost. In other words the content has been used for profitable purposes that do not benefit the content copyright owner or content license holder.

So what about local TV over cable? Twenty years ago I had cable but I couldn’t watch local TV without an antenna. Then one day, the CableCo said “you can add local TV for $5/month because we cut a deal with the local broadcasters”. That became so popular it was soon part of the baseline package. All their subscribers got local TV and the CableCo paid the local broadcasters. Aereo on the other hand, offers a low monthly fee to capture and broadcast local TV without paying anybody. The CableCo’s and the broadcasters didn’t like that, hence the lawsuit that went all the way to the SCOTUS.

People want to watch television content on their own terms, times and devices. There are solutions and there will be more choices, not fewer, as a result of this decision.

0. If you subscribe to cable this is a non-problem – Xifinity on Demand from Comcast – but that doesn’t count in this context.

1. The Major Networks – ABC, NBC, CBS, and FOX have free apps for iOS and Android smartphones and tablets and/or free and paid access through Hulu/Hulu Plus. Free means pre-roll and mid-roll ads but it only takes ~47 minutes to view a “one hour” show. Prime time is pre-recorded so why be a slave to a 10PM slot. Unfortunately many shows are delayed by hours or weeks.

2. Apps for AppleTV / GoogleTV / Roku / SmartTV

These can be ad-free under various subscription models. Google’s new Android L platform will significantly increase smart TV applications, surpassing proprietary approaches like Flingo. Apple is almost certain to provide similar incentives to both national and local broadcasters to create content apps for these devices.

3. Personal DVR and home-casting.
Content that arrives in your home legally can be archived and rebroadcast for you and your family’s personal entertainment. This is “fair use”. Slingbox is a perfect example of a legitimate solution. There are others, e.g. MythTV and XMBC are bring-your-own-hardware solutions. Boxee was a solution but they ran out of money after too many delays and pivots.

Aereo’s defeat in the Supreme Court will increase the solutions along these lines. The bill of materials for a personal DVR that connects to your ATSC antenna and your home router is less than $100. I would not be surprised if the backers of Aereo have hedged their bets in this direction. “But I don’t have an ATSC antenna (or I live in an area with no reception. Aereo was my only choice.” Not true. In the Bay Area, Comcast, DirecTV and ATT U-verse provide local TV. Some provide a local channel only service for about what Aereo charged. IThe difference is that these companies “paid” for the rights. Aereo did not. And It is likely that Google and/or Apple will also license access to affiliate or even national broadcasters to provide yet another alternative.

You can’t put the content behind a firewall and sell access any more than you can walk into a movie theater with a video camera or sell tickets to watch the World Cup on your big screen TV. The latter violation falls into the category of ridiculous to enforce, especially if your charge “beer”.

Read / Update my wiki article on Web to TV

Bitcoin Demystified (did I just pay $50 for a latte?)

bitcoin

Myth: Bitcoins are not as secure as US Dollars

Credit cards and financial account passwords are compromised everyday – whether by outright theft, hidden scanners or cameras that surreptitiously steal numbers, or hackers – causing millions of dollars in loss.

Counterfeit paper money is an ongoing problem. And if the bank robber asks for small bills at the teller window, he will likely get to spend them.

The Bitcoin “serial number” is registered and encrypted. Bitcoins can only be created solving complex cryptography problems in a competitive domain. It used to cost in electricity to mint a Bitcoin than they were worth. I doubt true now. And the cost to mine is dropping with electronic hardware (ASIC and FPGA) mining solutions.

Myth: Bitcoins markets are run by criminals and cater to criminals

If you don’t have the computer skills to mine Bitcoins, you must buy them through an exchange. Years ago, MtGox was the only option. Registration requires levels of identify verification as well as verification of funding sources. Trading on Mt. Gox is imperfect and randomly closed. On the other hand US stock markets have relatively narrow trading hours while BitCoins are traded 24×7.

Exchanges can be frozen or even shut down by government authorities. The use of Bitcoin markets to trade illegal goods or fund illegal activities is a valid concern. The FBI shut down TheSilkRoad and arrested the organizer. The news produced a brief 20% drop in BitCoin value, which became the buy opportunity of the year. The FBI isn’t going to dump Bitcoin accounts needed for a trial years in the future. The US government restricted Bitcoin usage on digital payment sites like Dwolla The Chinese government has made similar moves recently. On the other hand, a lot of people lost money when the US Government shut down online gambling sites, e.g. PokerStars and FullTiltPoker. Comments by government officials this week portend a positive environment for Bitcoin and other digital currencies. See Bloomberg news

A number of BitCoin exchanges and marketplaces have been funded in the last few months. Most of these are backed by Tier 1 venture capital funds with deep pockets and a long-term horizon.

  • BitInstant (New York): Platform for instantaneous Bitcoin transfers.
  • Bitpay (Atlanta): Lets customers of businesses ranging from software-makers to auto dealerships make payments in Bitcoin, then transfers cash to those merchants.
  • Coinbase (San Francisco): Makes a “virtual wallet” that lets users buy Bitcoin and pay for goods and services with it.
  • CoinLab (Seattle): Backed by Silicon Valley investor Tim Draper. Recently tried teaming with leading Bitcoin exchange Mt. Gox; the partners, however, are now in court.
  • Lamassu (New Hampshire): Has developed an ATM to instantly turn dollar bills into Bitcoin.
  • OpenCoin (San Francisco): Cofounded by E-Loan’s Chris Larsen, it has created a payment system to transact various currencies, including dollars, Bitcoin and a new alternative, Ripple.
  • Tradehill (San Francisco): Runs a Bitcoin exchange that competes with Britain’s Bitstamp, Russia’s BTCE and Tokyo-based Mt. Gox, among others.

If exchanges make you nervous, simply use them for conversions, then store your BitCoin in a personal secure, encrypted wallet. I use BitCoin-Qt.

Fact: Digital Money is here to stay

When is the last time you saw someone write a check in the grocery store, especially someone under 60? Today’s kids will shop with their smart gadget, not with a credit card and certainly not with folding money as my grandfather called it.

Bitcoin has advantages for micropayments (the smallest unit of payment is 1 Satoshi = 0.00000001 BTC) and that amount can be transferred at no cost. Independent artists could sell content for fractions of a penny without losing the $0.30 + 3% that Paypal charges.

Bitcoin is the first digital currency to gain a foothold. But it could be supplanted by something better. There is a gold analogy with respect to Bitcoin – the more that is mined, the harder it gets to mine more. Gold costs $300 per ounce to mine which sets a base value. Bitcoins mining essentially has half-life. About half of the maximum possible Bitcoins (21,000,000) has been mined. The remaining 11 million will be fully mined by 2017. This has several implications.

The US M1 money supply is ~$2.5T. In simple terms replacing M1 dollars with M1 Bitcoin equivalent would equate to a value per Bitcoin of $100,000. Most goods would be priced in mBitCoins or uBitCoins which sounds awkward. (1 MicroBitCoin = 10 Satoshi’s if you’re keeping score).

BitCoin software is open source. What if the US Treasury began minting DigitalCoins under a derived scheme, that combined US Government money management with the cyber protection advantages of BitCoin. Would Mt. Gox stand a chance?

Did I just pay $50 for a latte?

Success of Bitcoin hinges on the use of the digital currency for buying and selling legal goods and services. The price gyrations of the last few months make that a daunting challenge. Similar to the problems with runaway inflation, how much does an item really cost if the dollar basis is fluctuating by 30% to 90%. And that leads to the second problem – lack of merchants that accept BitCoin. At the end of the day, the rent has to be paid in dollars (or the local currency).

What is needed is a more efficient marketplace. Farmers hedge against price drops in crops on the Futures Market. Traders hedge again stock drops (or pops) with options. If people can make money regardless of the price movement direction, the forces that accelerate bubbles are diminished.

Full Disclosure

I gave my nephew a BitCoin for his 14th birthday. He sold it to buy an XBox One. I hope it doesn’t turn out to be a $50,000 game box.

SmartPhone, SmartTV, SmartEverything

Introduction

In only a few years, the Smartphone has become the personal computer of over one billion people. Another billion will be added in the next few years. The success is not merely due to the electronics in the phone, but the infrastructure of cloud services, applications and storage that allow personal computing devices to pervade and enhance our lives.

The technology inside today’s smartphones is comprised of ARM-based and Intel CPUs, graphics and video engines (GPUs), Mobile Interface IP, Memory components, peripherals, and wireless access devices (e.g. Cellular, Wifi, Bluetooth). Mobile Interface IP is governed by Standards, and optimized for low power consumption with high performance.

This same technology is now moving into a range of products beyond phones – powered by batteries or the grid – like wearable accessories, SmartTVs, SmartCars, SmartHomes, etc. These products will expand rapidly as the ecosystem evolves. The common denominator in these products is the same mobile computing Silicon IP that powers Smartphones. The AppleTV is just an iPhone without the cellular radio. The Samsung Chromebook has the same chipset as the Samsung Galaxy S3.

Read the full blog post on Arasan.com

Competition Spurs Performance in Mobile Storage

Many companies were surprised by the unexpected eMMC developments last November, when Samsung announced availability of a 260MB/s eMMC device. The announcement roiled the mobile storage industry and led JEDEC to hastily finalize a revision to the eMMC specification. The result is eMMC 5.0 with a peak transfer rate of 400 MB/s (equivalent to 3.2 Gbps). This moves eMMC slightly ahead of UFS v1.1 with 2.9 Gbps transfer rate. To further blur the advantage of UFS, some advanced features originally intended for UFS are now been planned for future eMMC revisions.The new eMMC 5.0 specification will also be introduced in the 3rd quarter of this year; its maximum data transfer rate will be 400MB/s or 3.2 Gbps in 8-bit bus operation. This high data transfer rate fills the performance gap before UFS entire ecosystem is available. In addition, an eMMC version of command queuing is now on eMMC roadmap for a future specification revision. It is also possible that eMMC will continue to increase the maximum data transfer by another factor of two.

Is UFS adoption in tablets and smart phones threatened by the recent advent of eMMC 5.0?

UFS version 2.0 will be introduced in the 3rd quarter of this year; its maximum data transfer rate can be 11.6 Gbps in a two-lane configuration with 5.8 Gbps per lane. In addition, native command queuing, inherited from the high performance SCSI protocol has been added. This provides multi-thread or multi-tasking operations which can provide an order of magnitude performance advantage over eMMC. A special extension has been added for Uniform Memory Access (UMA). This feature will allow non-volatile memory devices to extend on-die buffer storage for L2P (logical to physical) mapping. This can improve performance and device lifetime.

It seems like eMMC is positioned to fill the performance gap before UFS ecosystem (HW, SW, OS support, development tools, etc.) is ready. As a new technology, UFS will initially be adopted by high-end mobile devices. As the benefits of UFS are fully demonstrated in initial products, and the technology matures with increasing volume, UFS adoption will spread. UFS is supported by UFSA, a governing body of industry representatives chartered to define and certify compliance to the UFS standard.

UFS relies on the MIPI M-PHY. The M-PHY V2.0 supports transfers of up to 5.8Gbps (Gear 3) and is being adopted by additional standards bodies, e.g PCISIG for mobile PCIe, and USB for SSIC. This will further lower barriers to UFS adoption in high bandwidth, power critical applications.

Written by

State Machine Compiler: A Hidden Treasure

Most any digital system will require state machines to control the operation and implement algorithms. State machines are easily enough to write manually, but because they tend to require much iteration during development the frequent modifications can be tiresome and error prone. In addition, documenting the functionality becomes a chore as each iteration must be updated.

In the mid-nineties a solution appeared that seemed to address iterating and documenting state machines. A small company offered State CAD, a tool that enabled graphical entry of state machines. Along with the ease of graphical entry the tool compiled to RTL so the results could be synthesized to ant technology.

Graphical entry allows designers to see exactly what the logic does without having to pour over lines of code. Moreover, graphic entry is faster and less error prone than manual coding. Best of all, the graphic and code are always identical because one is the child of the other.

A few years after its introduction, the tool was acquired by Xilinx and added to the tool suite. It remained available without cost as part of the Xilinx suite and, best of all, technology independent.

The tool, however, disappeared after the ISE 10.1 release years ago for reasons that are obscure. For engineers who are familiar with the benefits of State CAD an internet search and a bit of luck will uncover its location. But those unfortunates who have never been introduced to the benefits of graphical entry may remain mired in their tedious ways.

While there are other tools available that have similar capabilities, none reputedly offer RTL outputs nor are they complementary.

The State CAD UI is straightforward and examples of its use abound. For example, see: http://ece.wpi.edu/~rjduck/state_cad_tutorial.pdf. In a nutshell, you select the circle icon for adding states and the arrow icon for transitions. You can add outputs to states and conditions to the transitions. There are countless other features.

The tool has a page for compile options that lets users select language (Verilog or VHDL) and other parameters such as to optimize for speed vs. size, or whether to register outputs. Errors show up at compile time and the tool offers suggestions for corrections.

Xilinx does not maintain State CAD and it is aging. One drawback in using it with Windows 7 in 64-bit mode causes a missing MSVC71.dll error. My attempts to circumvent the DLL problem were not fruitful.

Perhaps an enterprising Windows guru can give us a DLL workaround. Better yet, persuade Xilinx management to sell the tool for a dollar so that it may be made available for engineers to come. Otherwise, I fear its usage will in time will be swamped by operating system improvements.

Written by

Trends in Mobile Imaging

Introduction

There are four key elements to mobile imaging:

  • Media Processing
  • Display Technology
  • Content
  • Interface protocols between processor ICs and image controller IC’s

Today’s Application Processors for Smartphone and Tablet have substantial graphics processing capability. Most are fabricated in 28m high K metal gate CMOS nodes and the next generation of sub 20nm FINFETs will provide even more processing capability.
Display Technology for LED and AOMLED ranges from 200 PPI for large form factors to 440PPI for small form factors.
MIPI Interface standards dominate mobile electronics. Content is application and market segment driven. We’ll explore those in more detail below.

Smartphone Displays

HowToWatch4KSmartphone

Steve Jobs introduced the 326 PPI “retina display” of the iPhone 4 with the claim that it exceeded the discrimination of the human retina. According to Dr. Raymond Soneira of DisplayMate Technologies this was marketing hyperbole. He calculated that the maximum discernible PPI at 12 inches from your eyes is 477 pixels per inch. At 18 inches it falls to 318 PPI. I think most people tend closer to 18 inches so Mr. Jobs was basically correct. Perhaps we will be able to identify 4K Smartphones by their owner’s viewing style.

Advanced high resolution requires improved transistor density (and adequate dark space) and lower RC delays to reduce refresh times. There are two transistor technology candidates, amorphous oxide semiconductors and low temperature polysilicon. Faster refresh is required to improve motion blur. LCD TVs with 120Hz refresh are common and 240Hz is emerging. The same improvements will emerge in portable displays. Motion estimation and motion compensation (ME/MC) techniques use frame insertion to improve motion effects in action video and 3D gaming.[1]

Professional Tablets

Who will pay for more resolution than a digital movie requires? How about architects, general contractors, and a wide range of scientists and medical professionals.

High Resolution Content

ORPM

Cloud based telemedicine, i.e., image processing in the cloud for display on mobile devices, will play a key role in improving and distributing diagnosis. John Gore, director of Vanderbilt University Institute of Imaging Science (VUIIS) is researching medical imaging that produces extreme close-up views of the details of tissues, revealing the dynamics of cellular and molecular biology. [2]

Smartphones could be ideal for stereoscopic 3D technology allowing capture, playback and gaming. Codecs such as multiview H.264 (MVC), stereo JPEG (JPS) and JPEG Multi-Picture Object (MPO) are a critical element. OpenGL is used for gaming. 3D is achieved with dual images offset in color and phase, effectively doubling the resolution processing for a given display. Video frame rates are typically doubled to remove visual jitter. [3]

Image Interface Technology

The MIPI Alliance is a standards setting body with over 240 contributing member companies from the mobile platform, semiconductor and IP industries. Image display is supported by MIPI DSI with D-PHY connectivity. D-PHY allows power efficient, EMI reduced, serial communication between a CSI-2 or DSI host/device pair. 4 lanes of 1.5Gbps D-PHY supports displays of up to 2.5Kx2K resolution. A potential upgrade to M-PHY (DSI-2) could boost support to 4K.

Camera Interface Technology

Image capture is supported by the CSI-2 and D-PHY combination. Image data captured by the camera sensor should be presented to the CSI-2 Transmitter in RAW, RGB or YUV formats; the MIPI spec lists all the detailed formats that the CSI-2 connectivity infrastructure is required to support. The CSI-2 sensor interface provides you the option to compress the RAW data, and converts any kind of pixel data to bytes, which are then packetized and distributed over one or more lanes of the D-PHY. The higher the resolution of the captured image, the more speed and/or number of lanes needed. Given the max D-PHY throughput of 1.5Gbps per lane, with a maximum of 4 data lanes, the maximum camera resolution supported, assuming 24 bits/pixel in RGB format, with 30 frames/second is 8 megapixels. The proposed CSI-3 standard with MIPI M-PHY at Gear 3 (6.0Gbps) could support 60 fps.

What does the future hold

The trend to higher resolutions both in image capture and image display will continue. Will we ever see the quality of a RED EPIC Camera in a smartphone? Seems unlikely, but with stereoscopic multi-camera imaging and processing, who knows. One challenge that will not dissipate is the power required to transmit raw bandwidth between components in a system. From the tables above there is a level of resolution that will only be supported by compressed bit streams. Cameras will compress images on the fly and displays will decompress frames to buy 2 to 3 orders of headroom.

Technical References

  1. State of the art technologies and future prospective in display industry
    Moon, Joo-Tae, Samsung Display Labs, (IEDM), 2012
  2. Medical Imaging: Just What the Doctor (and the Researcher) Ordered: New Applications for Medical Imaging Technology, Mertz, L., Pulse 2013, IEEE
  3. Embedded stereoscopic 3D camera system for mobile platforms
    Aguirre, A.; Batur, A.U.; Hewes, G.; Pekkucuksen, I.; Venkatraman, N.; Ware, F.; Buyue Zhang, ICASSP 2012

 

Written by

The Law of Accelerating Returns for computing cycles

1943 – One ENIAC was ~300 Operations per Second (Ops)

2003 – no smartphones, no cloud, 500M PCs x 1Gops x 10% usage = 50*10E15 + BigIron

2013 – Cloud (Google, Amazon, Microsoft, Apple) computing cycles is ~200 petaflops {guess}
The sum of the Top500 supercomputers is >100petaflops
1B smartphone/pads x 1GOps x 25% activity = 250 PetaOps
PC sales have declined but still contribute >50*10E15
Assume Big Iron is constant (very conservative) [Note 1]
Total ~ 10**18 Ops

Ignoring the difference in FlOps and Ops.
In 70 years, Ops have increased by ~17 orders of magnitude, or about 2X Moore’s Law [Note 2]
In 10 years, Ops have increased by 200X, or about 2X Moore’s Law.

With (very fuzzy) activity assumptions, it’s not obvious that total computing is accelerating beyond Moore’s Law.

But total computing capacity appears to be accelerating.

[Note 1] IBM has 90% of the Big Iron market, sells ~4000 mainframes/year x 1GOps x 10 years of active installations = 4×10**13, so we can discount the impact since they’ve only gotten faster.

[Note 2] Moore’s Law is a metric of semiconductor performance which doubles every ~18 months.

See The Law of Accelerating Returns

Social Media and Search Dominate Ad Tools

  1. Marketing professionals would pay a company an average of $107,500 to manage an integrated digital campaign.
  2. Increasing sales (31%) and ROI (28%) are the goals most commonly cited as marketers’ most important for this year.
  3. Slightly more than half of respondents say that inefficient ROI tracking and measurement has hindered their success at work.
  4. Almost all respondents predict they will use online video this year. The most popular formats are in-banner (60%), in-stream (49%) and dynamic or customized video ads (42%).
  5. Roughly 2 in 5 respondents believe that they are behind the curve when it comes to digital marketing.

Are you behind the curve?

Written by

Benchmarking Mobile Storage

Introduction

NAND Flash is the nonvolatile memory behind SSD, mobile computing devices and mobile appliances (such as phones and cameras). Flash technology is the mainstream storage solution for mobile – and mobile is growing dramatically. Some analysts predict tablet shipments will eclipse PC shipments by 2016.

Why Benchmarking matters

Tablet computers and smartphones place increasing demands on NAND Flash device features – capacity, cost and bandwidth. But providing this requires a complex combination of electronic and software development. As a result benchmarking NAND Flash at the component and system level is critical to successful product design.

Benchmark testing spans a wide range of factors such as the device controller implementation, the host interface standard (SD, eMMC, UFS), type of NAND technology (SLC, MLC, TLC), device drivers and file systems and the application requirements, e.g. sequential, random, read-write, etc.

NAND Flash Technology is Tricky

Amplification

NAND Flash cells are read at the page level and erased at the block level. Cells must be erased before programmed. For example, a 16Gb NAND Flash from Micron [1], uses a 4KB page and a 512KB block. Every time data is written to a block the entire block is erased. Internally valid is copied to available blocks. In effect an external write will produce multiple internal writes.

Block Management

Block management accomplishes several tasks related to translation of logical addresses to physical addresses including bad block tagging, dynamic buffering for performance, and garbage collection. This function may be implemented in a Flash Translation Layer (FTL) in firmware or a Flash File System (FLS) running on the host.

Wear Leveling

Wear leveling manages the local to physical mapping so that erase/write cycles are distrusted evenly over the entire range of the memory array to optimize the life of the device. Typically this is implemented in controller FTL where the P/E cycle count is known. New approaches based on BER optimization are emerging. [2]

Error Recovery

Some error recovery can be implemented in the controller command logic (with extra P/E cycles) for read / write disturb errors when adjacent (unintended) cells voltage levels shift from I/O activity. Other errors such as voltage shift from alpha particles injection rely on ECC to resolve.

ECC

Error correction code (ECC) solutions are based on Hamming, Reed-Solomon, BCH [3] and LDPC. Hardware solutions scale in complexity with the NAND block size and the error correction requirements. SInce soft errors scale with shrinking NAND process geometry, the hardware costs of ECC are increasing. Low Parity Density Codes, use statistical algorithms implemented in firmware to reduce block-level BER with less integrated circuit area. [4]

Signal Processing Optimization

Beyond merely correcting errors, signal processing recovers correct bits by analyzing multiple read voltages. Integrating signal processing, ECC and FTL into the device controller, e.g. “managed NAND” by Anobit [5]. NAND device suppliers, e.g. SK Hynix and Micron, are integrating control functionality like ECC and signal processing onto the NAND die or dedicated controller die in the NAND package. [6,21]

Implications

Controller design is intimately tied to the NAND device. The degree is only increasing as NAND technology shrinks and NAND devices grow in complexity and density. Mobile Storage devices may exhibit dramatically different results depending on price and application requirements.

The full white paper can be read here

Written by