ARM – Arm Holdings Takeover and Open CPUs

In the financial news was the buyout of Arm Holdings by Softbank. Arm Holdings are the British folks who design all the ARM CPU chips. Like the ones used in the SOC System On Chip that’s the heart of the Raspberry Pi (along with oodles of network routers, embedded systems, cell phones, and gadgets all over the world.)

There was about a 47% price bump in the offer over the present market price. Now part of that came out of the buyer being Japanese and the ARM shares being priced in £ Sterling (it being down lately) but not the whole jump. A large part was the buyer (Softbank) paying up. Why pay up?

Well, Intel makes very good CPUs, but they are very large, suck lots of power, and cost a fortune. ARM chips are very small, sip barely any power, are dirt chip to fabricate, and are in just about everything that isn’t a PC.

Softbank has been playing the “rollup” game, buying all sorts of telecom related properties, and “growing” like a weed. Not too long ago they bought up Sprint / Nextel. I put growing in quotes as rollups LOOK like they are growing on financial reports, but just because you glue two $Billion companies together it doesn’t mean you actually created a $Billion of something new. Often “rollups” end badly. (See Rite Aid as an example, or Tut Systems, or a dozen others). It takes a fair amount of digging to determine if the rollup is “strategic acquisition” or just “hide the stagnation”. Beware the “synergy story”…

Here’s what Arm Holdings did on the news (and I don’t know what currency this is in or what the conversion was):

ARM Arm Holdings on Softbank Takover

ARM Arm Holdings on Softbank Takover

The first thing to notice is the rocket up at the open on the news of a high premium buyout. The second thing is the folks selling on higher than normal volume after that to “lock in the gains” in case the deal falls through.

Generally, price rockets up to near the offered buyout price. Not exactly to it, as there is a risk of failure fo the deal. Most folks ought to “sell the news” if the price is at all near the takeout offer. If it isn’t near, someone is expecting the deal to likely fail. Sometimes the offer is near the prior market price, then the bump may be to at, or above, the offer as “folks in the know” expect a higher bid or a bidding war. Then the better strategy is usually to hold out for more. Each deal is unique, though.

So what is Arm Holdings?

The short form is that they are a British CPU designing company that makes a nice, efficient, and very widely used design for a CPU chip.

Generally speaking, someone else makes all the chips. This matters a lot.

So what Arm does is license their IP (Intellectual Property) to others to build. Folks like Broadcom and Allwinner. Though even in those cases they may not actually fab their designs, but just glue the Arm design onto other bits and send that collective IP to a fab shop…

Allwinner Technology is a Chinese fabless semiconductor company that designs mixed-signal systems on a chip (SoC). The company is headquartered in Zhuhai, Guangdong It has a sales and technical support office in Shenzhen, Guangdong, and logistics operations in Hong Kong.

Since its founding in 2007, Allwinner has released more than fifteen SoC processors that have been widely implemented in Android-based tablets, as well as in smartphones, over-the-air OTT boxes, video camera systems, car DVRs, and car media players.

The company provides a backdoored Linux kernel, that is available in their products, lowering their security considerably. The company itself has also released low-quality contributions to the open-source community, via the release of mainly binary blobs, which is in clear violation of the GPL license that the Linux kernel uses.

Why I prefer the Broadcom chips to the Allwinner chips…

But notice that you don’t need a “CPU Designer” to make an integrated SOC (System On Chip) cell phone? Just license the one from Arm, glue it to your interface designs, and send it to fab. Oh, and since Linux already runs on ARM Cores, a port to your new system is pretty easy and quick. Mostly a few drivers and some debugging. And maybe a backdoor for The Party Bosses if you are in China…

ARM came to this position by having a pretty good chip design early on, and being willing to license it. In the 1980s, they were NOT that unique. Lots of players made CPUs. But they mostly wanted to fab their own… so all those “Me Too!” folks licensed from Arm. I was at Apple then, and my desk was not far away from the guy working on the early PDA (Personal Digital Assistant) product ideas. What later became the Newton, the Pocket Crystal (General Magic spin out), and eventually the iPhone – after a generation or two. Apple was trying to decide “make our own or buy in?”. He talked to Arm, looked over their stuff, and pitched it as the best solution for lowest cost. A hundred other shops were busy doing the same thing…

So National Semiconductor had a nice 32 bit chip. Now forgotten.

MIPS had a great chip set. Barely remembered by a few.

DEC made the PDP into a chip set. Dust of time.

Apple used the ARM chip for the low end, but decided to design a Super Duper Chip for the future desktop. That was the project I was on, running the hardware used in the designing. That Cray time. Though that project was killed (in a very stupid way, IMHO) the chip design went on to be the foundation for the Power PC (merged with the IBM RS 6000 IIRC) via Taligent corp (an IBM / Apple combine). Apple used that chipset until recently when they went over to the Intel chips (that had bought tech from HP as HP left the CPU design business and my Old College Roomie who worked at HP went on to other things…) As you can see, that era of my career was “in the thick of it” on chips and chipsets.

The “bottom line” is that then, to design and debug and get to the point of fab, a whole new CPU chip, could run into the many $MILLIONS per chip. Then, when done, you get to start all over on the next generation. A LOT easier to pay $1 / chip (now closer to 5 ¢ / chip in bulk) to Arm for a tested design…

So that’s what Softbank is looking to buy. A “strategic position’ in telecom infrastructure IP owning the designs used in much of the telecom gear globally.

Who is Softbank?

You can “hit the wiki” for the details. I’m just going to quote a couple of bits for illustration.

SoftBank Corp. (Sofutobanku Kabushikigaisha?), previously as SoftBank Mobile Corp., Vodafone K.K. (also known as Vodafone Japan) and J-PHONE is the telecommunications subsidiary of SoftBank Group, including mobile and fixed-line services.

Masayoshi Son is chairman and Ken Miyauchi is President and CEO.

SoftBank store, Hankyu-Ibaraki, Futabacho Ibaraki-City Osaka, Japan
SoftBank Mobile operates W-CDMA (UMTS 3G) network (“SoftBank 3G”). SoftBank’s 3G network is compatible with UMTS and supports transparent global roaming for existing UMTS subscribers from other countries. SoftBank 4G uses TD-LTE / LTE. SoftBank offers 4G speeds of more than 110 Mbit/s. SoftBank WiFi Spots are available almost everywhere in Japan.


J-PHONE store in Nagoya

The company was originally founded in 1981 as the mobile phone division of Japan Telecom under the name Digital Phone. J-PHONE Co., Ltd. was formed in 1999 by the merging of Digital Phone Group (DPG, 3 local companies) and Digital TU-KA Group (DTG, 6 local companies, not to be confused with TU-KA). Japan Telecom owned a stake of 45.1%. In October 2001, the British mobile phone group Vodafone increased its share to 66.7% of Japan Telecom and 69.7% of J-Phone. On October 1, 2003, the name of the company and the service brand was officially changed to Vodafone. The growth and success of the company during this period is due in large part to then president Bill Morrow.

On March 17, 2006, Vodafone Group announced it had agreed to sell its holding of Vodafone Japan (Vodafone K.K.) to SoftBank for about 1.75 trillion Japanese yen (approximately US$15.1 billion). On April 14, 2006, SoftBank and Vodafone K. K. jointly announced, that the name of the company will be changed to a “new, easy-to-understand and familiar” company name and brand.

It was announced in a press conference on May 18, 2006 that the new name would be “SoftBank Mobile Corp.”, effective October 1, 2006. SoftBank started the rebranding around June 14, 2006.

Or is it the parent, SoftBank Group Corp? (Welcome to the world of Japanese Keiretsu and layered holdings / cross holdings)

SoftBank Group Corp. (Sofutobanku Gurūpu Kabushiki-gaisha?) is a Japanese multinational telecommunications and Internet corporation, with operations in broadband, fixed-line telecommunications, e-commerce, Internet, technology services, finance, media and marketing, semiconductor design, and other businesses.

SoftBank was established in 1981 and is headquartered in Tokyo. As of 2015, SoftBank was the 62nd largest company in the world (based on a composite of sales, profit, assets and market value).

and on and on.

Basically it’s a fairly large telecoms corp going for the usual Japanese vertical integration. From chips and wires to telco service to the end customer.

In such a behemoth, a tiny little boutique CPU designer will be rapidly lost. THE big questions are:

1) Can they keep the Arm Holdings culture alive and thriving?

2) Can they keep the present employees and attract similar folks in the future?

3) Will they be bright enough to keep the present business (licensing) model?

4) Will they be so self centered as to give themselves favor in chip availability?

If they fail in any of those, the end result will be all the other telecom players leaving the ARM chip for other designs. Sooner, or later.

Arm Holdings was never seen as a direct competitor to their licensees. Broadcom, as a random example, doesn’t have to fear Arm Holdings giving preferential access (in time or dollars) to Allwinner. BUT, if Allwinner gets a design 6 months earlier, or at 1/3 off the cost; Broadcom would be strongly disadvantaged on time to market or parts cost. At that point, they MUST change vendors.

Over time, if Japanese Conglomerate Arm isn’t able to attract the kind of guy who works at Arm Holdings, the product will become stale and folks will just ‘move on’ to other providers.

Who Are The Alternatives?

Well, the one that comes to mind first is MIPS. Now “Imagination Technologies”.

The “Mips chip” has always been a good one. I was a bit disappointed when ARM beat them in the small scale licensing market. They do have enough market presence (mostly in bigger machines) to have Linux ports available.

MIPS64 I-Class I6400
Multiprocessor Core

MIPS Series6 Warrior I-class core offers 64-bits, simultaneous multi-threading
SIMD, virtualization and more
Classic Cores

Widely licensed and cost-effective solutions for embedded and multimedia applications.

Though I think they need to get the idea of “Classic” (that seems dangerously close to “legacy” to me…) out of their mind and instead focus on “power efficient embedded” cores… But then I’m not in marketing, likely for a reason ;-)

For my money, it is a “first stop” for things like tablets and cell phones, especially as performance demands increase.

Imagination Technologies Group plc (LSE: IMG) is a British-based technology R&D company, focussing on semiconductor and related intellectual property licensing. It is most noted for its PowerVR mobile graphics processors, MIPS embedded microprocessors and for its Pure consumer electronics division. It also supplies radio baseband processing, networking, DSP, video and audio hardware, VoIP software solutions, cloud hosting and silicon and system design services. The company is listed on the London Stock Exchange.

Note that PowerVR. In some SOC designs, you get an ARM CPU core and a PowerVR graphics core, each licensed from a different company. It would be a small ‘leap’ to go from that to an all-MIPS license…

Then there is an interesting design from another company with an extensible CPU core…

Tensilica is a company based in Silicon Valley in the semiconductor intellectual property core business. It is now a part of Cadence Design Systems. Its dataplane processors (DPUs) bring together the strengths of CPUs and DSPs and custom logic with 10 to 100 times the performance[citation needed], making them suited for data-intensive processing tasks.

The Tensilica brand is best known for its customizable microprocessor core, the Xtensa configurable processor.
Other products include: HiFi audio/voice DSPs with a software library of over 125 codecs from Cadence and over 55 software partners; IVP Image/Video DSP, designed to handle complex algorithms in imaging, video and computer vision; and ConnX family of baseband DSPs ranging from the dual-MAC ConnX D2 to the 64-MAC ConnX BBE64EP.

Tensilica was founded in 1997 by Chris Rowen (one of the founders of MIPS Technologies) and was initially staffed by former employees of several other Silicon Valley processor and electronic design automation companies. It employed Earl Killian, who contributed to the MIPS instruction set, as chief software architect for several years. On March 11, 2013, Cadence Design Systems announced its intent to buy Tensilica for approximately $380 million in cash. Cadence completed the acquisition in April 2013, with a cash outlay at closing of approximately $326 million. [3]

Being owned by Cadence is both good and bad. Folks using Cadence will find it a benefit. Folks committed to their competition may see it as an annoyance. Yet the cores are available.

Then there’s. ARC International, now owned by Synopsys. (Notice a trend here of being bought?)

ARC International PLC was the designer of ARC (Argonaut RISC Core) embedded processors, which were widely used in SoC devices for IoT, storage, digital home, mobile, and automotive applications. ARC processors have been licensed by more than 200 companies and are shipped in more than 1.5 Billion products per year. ARC International was acquired by Synopsys in 2010.

The roots of ARC International date back to the early 1990s. The company was founded by Jez San to build upon the 3D accelerator technology previously developed for the Super Nintendo Entertainment System by a division of Argonaut Software. This forerunner to the ARC was originally called the Mario (Mathematical, Argonaut, Rotation & I/O) chip and later dubbed the Super FX. It went on to sell millions, at the time outselling ARM or any other RISC core.

Following the success of the Super FX, its designers were split from the main company to a subsidiary called Argonaut Technology Ltd (ATL). The design was renamed to ARC and marketed as a general-purpose configurable microprocessor. Later, ATL spun off as a separate company, ARC International. In 1995 Bob Terwilliger took over as ARC’s first CEO. He created the company licensing strategy, commercialized the product including the acquisition of Metaware, VAutomation and Precise Software. He raised $50 million pre-IPO and took the company public in September 2000, raising an additional $250 million.

A Moment To Ponder

Now take just a moment to look at those stories. A few guys with an idea found a company and design a chip. They sell the idea and the products, get noticed, and get bought up. Wash and repeat. It doesn’t take much to start a new chip design firm. A half dozen guys with clue and a bit of money.

Buying ARM, does not and can not have any monopoly power in the market. ARM has position power as it is widely used; but a truth of the electronics industry is that you are always one product design cycle away from being designed out. At one time, National Semi made a hot CPU chip. Now? I don’t know, but haven’t heard news on them for years so likely not. Maybe? Texas Instruments was “hot” for a while. Now? Who knows. Intel makes the hot Desktop CPU (with AMD constantly nipping at them…). Intel also makes a variety of embedded CPUs (though for licensing CPU Core designs for your own integration and fab, not such a good choice, as Intel is also your competitor). There are loads more, too. I can’t begin to list them all.

So between the “few guys with an idea” and the semiconductor designers (from Giant Global Blob to Boutique Shop) you have a wide range of options, if the ARM deal makes ARM less interesting.

But Wait! Theirs More! Annnndddd… It’s Free!

Think of it as “Open Source Meets Hardware and Falls In Love”…

They have a LOT of different hardware designs ready for “tape out” (just before fab). All free. Some of them of known decent designs. Others oddball, but interesting. A few just an idea, for now. Then my favorite, the “RetroComputing” cores. Folks have gone back and implemented retro instruction sets in modern design systems.

In many cases they have these cores running in FPGA (Field Programable Gate Array) configurable hardware chips to prove and test the design. This one is an ARM 7 compatible core:,storm_core

(You do need to get an account and log in to download the layout or view the statistics)

I personally like the PDP 11/70:,w11

The project contains a complete DEC PDP-11 system: a PDP-11/70 CPU with memory management unit, but without floating point unit, a complete set of mass storage peripherals (RK11/RK05, RL11/RL02, RK70/RP06, TM11/TU10) and a basic set of UNIBUS peripherals (DL11, LP11, PC11), and last but not least a cache and memory controllers for SRAM and PSRAM. The design is FPGA proven, runs currently on Digilent Arty , Basys3 , Nexys4 , Nexys3 , Nexys2 and S3board boards and boots 5th Edition UNIX and 2.11BSD UNIX.

This is a retrocomputing project, rebuilding hardware from the late 70s and running historical software. To get into the tune see Figure D-1, a 11/70 console, and Figure F-2, a baseline system setup.

News 2016-06-26: Release w11a_v0.73 available for details see section Releases.

News 2014-06-19: doxygen generated code browsing available for vhdl and C++ sources. No documentation text added so far, but helpful to navigate through the code. Good starting points are vhdl module list, or sys_w11a_n3 source, or C++ class list .

Then again, that was one of the first computers I ever used… I miss toggle switches on the front of computers..

Or how about that old National Semiconductor 32 bit CPU that was so “hot” once?:,m32632

M32632 is an implementation of the Series 32000 architecture of National Semiconductor. This 32-bit architecture was popular in the 1980’s and began to disappear in the beginning of the 1990’s. The first microprocessor was the NS32016. The third generation CPU was the NS32532. This processor is the basis of M32632. In addition M32632 implements the functionality of the NS32381 floating point unit.

But, for folks hopelessly in love with the new, how about a Java optimized core?,jop


– Very small core:
– about 2000 LCs – 3000 LCs (configurable)
– fmax is 100 MHz in a Cyclone EP1C6
– Real-time features:
– architecture designed to simplify WCET analysis
– cycle accurate time interrupt (not tick based)
– real-time enhanced thread model
– WISHBONE master


JOP is the implementation of the Java Virtual Machine (JVM) as concrete machine in hardware. The design has been sucessfully implemented in low cost FPGA devices from Altera (ACEX 1K50, Cyclone) and Xilinx (Spartan II and Spartan-3).

JOP is open-source under the GNU General Public License, version 3.

The point? IFF every single chip design house on the planet gets bought up by the various players, you can always just “roll your own” via an open source design. What used to take $Million of scalar processor to “reduce to silicon” can now be done on a home PC. What used to take a dozen guys with skilz working for a year or two can be downloaded for free. What used to take a few $Million of Fab to make, can be run in a FPGA. (For volume you can call up a contract fab company and hand them the tape…)

In short, where CPU design and build WERE terribly arcane, expensive, and hard things; they are now college Masters Thesis projects (or MIT undergrads, I’m sure ;-) and cost is approaching free. Not really all that hard either, with modern CAD software.

In Conclusion

I’m quite certain that across the Silicon World there are dozens of designers of cell phones, tablets, routers, telecom gear, embedded systems of all kinds; who are looking at their technology road-map and where it says “ARM 64 bit model TBD” penciling in “Or competitor chip TBD, pending review”.

For the next couple of years, nothing will change. Licenses are in hand, designs set, chips in fab. It’s the products 4 to 8 years out where change will hit. IF Softbank has a kind of “hands off just keep doing it” approach to Arm, it will retain dominance. However, any ‘bump in the night’, there are loads of alternatives.

Some folks will shift designs just because they fear a strong vertically integrated competitor owning their IP source and potentially knowing something of their plans (if YOU made a home telco IP router would YOU want a major supplier of them to know you wanted a license to make 2 Million A15 cores next year?). There WILL be ripples through the whole industry, short term.

Long term, it depends on Softbank and what they do with Arm… and that half dozen guys in a Venture Cap office making their pitch last week…

Subscribe to feed

About E.M.Smith

A technical managerial sort interested in things from Stonehenge to computer science. My present "hot buttons' are the mythology of Climate Change and ancient metrology; but things change...
This entry was posted in Economics - Trading - and Money, Tech Bits and tagged , , , . Bookmark the permalink.

14 Responses to ARM – Arm Holdings Takeover and Open CPUs

  1. M Simon says:

    The way ARM chips handle stacks is an abomination. But I do use them in my designs.

    I also use STM ARM chips (not listed on the blog). For power sipping I prefer the NXP chips.

  2. E.M.Smith says:

    @M. Simon:

    Feel free to post links / summary of any other interesting chips that I left out. Like I said, I didn’t have time enough to sort through all the options and figure out the better ones… but if someone has a “known good one”, by all means, promo it!


    I likely ought to have also mentioned that the “opencores” folks have a single chip design they are pushing forward as an Open RISC design with lots of support (not just some guy’s masters thesis idea…)

    The OpenRISK 1000, complete with virtual system implementation, tool chain, the lot. So even if you don’t want to buy a FPGA and build real hardware, you can still get the world ready for the actual hardware…

    Just the TOC:


    1 Overview of the OpenRISC 1000 project
    1.1 Architecture specification
    1.2 Licensing
    1.3 Processor Implementations
    1.4 SoCs based on OpenRISC
    1.5 Toolchains
    1.6 Operating systems
    1.7 Bootloaders
    1.8 Conferences and meetings
    2 Get source code
    2.1 Ubuntu OpenRISC VirtualBox image
    2.2 Repositories for latest versions
    2.2.1 GitHub
    2.2.2 OpenCores SVN repository
    3 Bug reporting
    3.1 Enabling Bugzilla
    3.2 Reporting a bug
    3.3 Bug work flow
    3.4 The old bugtracker
    4 Guide for developers
    4.1 Contributor guidelines
    4.2 Project maintainers
    4.3 Future work
    5 Further information
    5.1 News
    5.2 Manuals
    5.3 Tutorials
    5.4 FAQ
    5.5 Mailing lists
    5.6 OpenRISC forum
    5.7 IRC

    Note that Ubuntu OpenRISC VirtualBox image… So you can run Ubuntu on a virtual machine on a virtual chip on just about anything…

    Even your PC with Micro$oft on it… ( I’ve run Virtual Box Linux on a PC… while working at a site “unnamed to protect the innocent”… I brought up a virtual Sun Box on a PC and played a bit with “big endiean” Unix / GIStemp. Got it to run, but man it was slow… )

    Also note the SOC tab. They have a full system on chip layout for your FPGA… IF you wanted to, you could do a ‘tape out’ and hand that to a fab house (along with some money…) and have a custom chip built that already has Ubuntu running on it… Just saying… if you know any really rich geeks looking for a bit of fun… Think of a R.Pi like board with a totally open source stack from hardware to firmware to Linux… I’d buy 2 in my first order…

  3. M Simon says:

    National Semiconductor is now owned by TI. The TI MSP430 series is rather interesting. It is slow but has built in FRAM on the latest designs.

  4. M Simon says:

    I’m looking to create a RTX2000 design (upgraded to 32 bits) I plan to buy all the ancillary IP (I2C interface, FLASH etc.) and get the chip done at an open FAB. I REALLY like FORTH. It cuts development time by 3X or better compared to “C”. It has none of the built in “protection” C does. On the other hand bits are bits. No casting or other such abominations required.

    Designing your own PCBs? I really like these guys.

  5. Ahhh – Forth. That explains your disdain for the ARM stack implementation :)

    I remember experimenting with that back in the day on a VAX-11/780…

  6. Nick F. says:

    Great survey. Caught me up from 20 years ago when I kept an eye on on this for my employer.

  7. cdquarles says:

    Funny you mention this. I’ve been playing with VMs. My latest foray failed. Back to VirtualBox I go. Let me see if that works.

  8. cdquarles says:

    Heh, Forth. I played with it in the 80s and 90s. Forth is an interesting language, but RPN was not my thing. Sure, I could have grokked it eventually, but my TI beat my HP calculator back then. ;)

  9. Regis Llanfar says:

    I would say that a good Forth programmer could out-metal a good C programmer due to the need to be intimate with the stack.

    Then again, I’ve had too many Guinness…

  10. E.M.Smith says:

    Well, my bias is more mixed. I like threaded interpreted languages (like Forth) but usually use them via shell scripts (shells are threaded interpreted). I also llike RPN having used an HP 35 through college after the slipstick in highschool and first year college). Then my favorite machine was the then exotic dual processor stack machine, the B6700 from Burroughs.

    But by the time I ran into Forth, I was up to my eyeballs in Unix and C then Python and Perl were the big thing doing sysadmin… so never got to do much past the intro class.

    Maybe someday I’ll have time to get good at it…

  11. E.M.Smith says:


    Interesting chip…

    Sometimes I think it would be fun to get the “Burroughs on a chip” and make a stack system of my own… then I look at all work involved and decide a Pi is fine ;-)

    Besides, they phased out the stack chips in 2014 so I’d have to buy an old recycle box…

  12. E.M.Smith says:

    32 pages on MIPS based SOC makers and their products:

    Yes, I was scouting the future landscape possibilities, like hundreds of others :-)


    You know the MIPS has stack pointer and frame pointer and….

    Ought to work well with forth.

  13. cdquarles says:

    Speaking of stack and frame pointers, lots of Intel x86 designs had segment registers. I’d call those a form of frame pointers, too. Still, being used to MOSTEK/Western Digital 65×02 and later processors at the time, I couldn’t wrap my head around the segment registers. Hmm, I suspect that the later IA32/AMD64 designs still have segment registers but force them to point to a fixed address.

  14. Nick Fiekowsky says:

    Great article on ARM in data centers.

    James Hamilton is with Amazon, has lots for informative blog pieces. At Amazon scale, processor economics are very important. One can imagine that technology vendors, from parts through complete units, are extremely sensitive to Amazon’s direction.

Comments are closed.