"Transmute" - Design Philosopies

I realized that I had forgotten to describe a very, very important step: Philosophy. In this case I am using the word incorrectly. It is actually World View that I mean. From Wikipedia:

"A comprehensive world view (or worldview) is the fundamental cognitive orientation of an individual or society encompassing natural philosophy, fundamental existential and normative postulates or themes, values, emotions, and ethics."

Now what has this to do with chip design? Well you might ask! In this case it is the fundamental underpinnings and assumptions that surround the design.

The world view is formed from facts and opinions and our emotional reaction to them.

So here is the list of suitable facts for the "Transmute" project:

• We have only built cell based custom digital cores
• None of our typical projects are as large as this
• The design flows are not stable - considerable development will be required
• Several custom digital modules will be required to interfaced to licensed IP
• We can experiment with power saving strategies
• We are very experienced with custom analogue blocks
  

What, as a designer, is my emotional reaction to this? Simply this: Excitement and Concern. There are a lot of unknowns.

Hence our design World View (although a different design will have a different world view):

• Digitally Conservative
• Analogue Progressive

One you articulate this World View clearly you realize that this is the guide for your choices, in our case:

• Hard IP: Eliminates the risk from core synthesis
• Digital part to be separable from the analogue part so we can drive them separately. This will minimize the cost in case of failure or error - we would get at least one working part from the chip
• Digital power saving strategies to be applied only after a high confidence in non-power saving digital structures obtained
• It must be possible to disable the digital power saving structures if they are added
• Flow decisions should favour tools which we have experience with
  

These decisions then guide the rest of the design flow. Also new choices can be matched against the worldview ensuring consistency.

Just bear in mind, your world view for this design may need modifying as the project progresses however if it does then you need to reconsider most of the work you have already carried out. This is typically far more important if your original plan was aggressive and you move to conservative - check work and assumptions already made!

The next SoC design - working title "Transmute" begins

I have a fun new hobby - designing the digital part of a new, mixed signal SoC which we are building here. It is my intention to blog the full design flow discussing the various challenges and issues as we encounter them.

Design Basics

• Foundry
• Technology
• IP
• Flow
  

The importance of these choices of these depends on the following thoughts:

• What have we done before
• Who have we worked with before
• What have we got that works
• What we need to do and what we would like to do
  

Planning a chip begins after you decide the basics above, i.e. foundry, technology and the needed parts and finally tool support. Sometimes in industry you have the luxury of a couple of other choices e.g.:

• Cell library (we use the ST supplied one but there are others, Faraday for example)

Basic Overview Spec

1. Two full custom ADCs for testing and evaluation
2. A single core SoC to test our mixed signal integration and apply tests
   
3. Simple connectivity to the SoC

Initial Decisions

In our case after some mulling over our experiences we decided on:

1. ST Microelectronics via the Circuits Multi-Projets MPW broker
2. Their HCMOS9 130nm process
3. IP Blocks (see below)
   
4. The flow, more on that later

 IP Blocks

As a university our access to IP is somewhat limited. Making a SoC requires quite a few modules like memory controllers etc.

However to the rescue rides Synopsys' DesignWare Library. It includes a basic selection of AMBA AHB/APB bus connected peripheral set and a memory controller which supports SDRAM and Flash.

To drive the the DesignWare library is the coreTools GUI (which is surprisingly hard to find) which in a nice graphical environment enables you to construct AMBA bus structures containing DesignWare peripherals. So this drives our choice of some of the tools (coreTools and DesignCompiler Ultra) as well as their DesignWare basic IP.

Onto a much more interesting question. What core to use? Considering our other IP is AMBA an ARM is the most likely choice however there are others we might use:

• The LEON3 from Aeroflex is a Opensource (GPL), VHDL 32 bit SPARCv8 CPU with an AMBA bus. It is, however, very much bound to its peripheral library and autoconfiguration so would take a lot of work. It would also need to be customised to use the HCMOS9 SRAM blocks
• The IBM PPC 405 is a 32 bit PowerPC core designed for embedding which is licensed to universities for teaching and education. It is the distribution created for the Synopsys flow which allows easy integration with AMBA peripherals. It is a softcore, i.e. it would need to be synthesised from RTL

In this case we wish to reduce our risk and design effort. Licensable from CMP is a foundry guaranteed ARM946E-S hard IP (i.e. they have already synthesized it and checked it) on their 130nm process. So we are going ahead with DesignWare soft IP surrounding a hard IP ARM946E-S. The benefits of hard IP are that ST have de-risked the design for us.

The features of the ARM946E-S are:

• Excellent compiler/OS/application support. ARM is one of the best supported architectures in both Opensource and commercial software
• Small and quick, approximately 3mm² and clocking at 200MHz (both of these are significantly derated estimates at this stage of the design
• MMU allowing full OS support
  

In addition we need the following types of IP:

• A PLL - this is pretty much required for any SoC these days (available from CMP)
• Level shifters and isolators - needed to try advanced power saving strategies in digital logic

The current initial design of the digital segment of our chip is best shown in a block diagram below:

We will talk of flows and the analogue integration later.....

The sheer size of an old Sun "midrange" server

This one is going back a bit but I have obtained it for legacy support. Its a Sun V880 with 6x UltraSPARC III 750MHz processors and 12GB of RAM. Still a computer to be reckoned with [Update: Spent $50 on a new CPU/RAM board to take it to 8x CPU and 16GB RAM. eBay is great for this kind of thing]

Its as high as my desk, as large as a fridge lying on its side and sounds like a very loud aircon. [Update: When it is running under an OS at low load it throttles the fans and becomes quite quiet]

This one was going to be built with Solaris 10 on a software mirrored root using UFS but the new spin of Solaris 10 allows installation and booting off ZFS. A much more sensible choice when I have 6x 36GB disks.

Modding an HP Proliant DL585 to take an EVE Zebu UF2

I am the proud "controller" of an EVE Zebu UF2 emulation accelerator (a big PCI card covered in FPGAs). One of its biggest selling points is their Transactors which allow you to present your design with the interfaces it would experience if it were real, completely abstracting away the PCI bus.

However when I say big I mean triple height and full length. It uses 64 bit PCI/66. (For those considering purchasing a UF4 card most PCs can't fit it as it is larger still).

So I need a computer to host it in!

Enter the HP DL585 - a lovely machine. Built as solidly as a tank and everything is terribly easily accessible either from the front or the top. Everything. At a push I think I could swap out the motherboard and be back up and running in 30 minutes from cracking the case.

The original engineers designed it for cable-less operation (and very nearly succeeded!) but the EVE ZeBu-UF needs power from a 4 pin molex (found on most pre-SATA HDDs).

So what to do? Spotting this SCA backplane and knowing that HDDs take 12V, 5V and 3.3V power I had my solution. Mod a molex connector onto the pins on the back of one of the SCA sockets.

Firstly I checked the specifications for the ZeBu: 25W at 12V which is a current of 2.08A. Secondly I checked the specifications of the SCA2 connector: Section 6.4.1 of the SFF-8046 standard specifies a surge current of 5A and a constant supply of 2.5A. This means that one socket can supply the necessary power.

However, several provisos first: This will void your warranty. The SCA-2 connector transfers power to the drive using several pins for each voltage (12V, 5V and 3.3V) to reduce their resistivity to high current. By design the clusters of pins supplying the same voltage are next to each other making it easier. Probe for the resistance between the pins you think are the right ones - it needs to be zero.

The mod requires soldering the little clusters of pins together with fine wire (get some strands out of a multistrand wire) and then attached the cables. Note - don't include the Precharge pins - wiring them to the rest of the supply could cause unexpected operation as they will no longer function correctly when you insert a HDD in the front. After you are done, check, double check and triple check. Use a magnifier/microscope to check the soldering. Clean the area of any bits of solder. Probe the resistivity again. This could be an expensive mistake if there is a problem!

Power on (fingers crossed) and bingo - 12V and 5V.

Here is a photo of the modded card in position:


...and a photo of the UF2 in the PCI bay (I have two power connectors just in case I can get another UF2 - wishful thinking I know):

Loading new maps into Nokia Maps on an N95 without a cable or bluetooth

Foolishly I arrived in San Francisco for MSE without loading the California maps into my phone.

This presented me with a pretty conundrum. Nokia's recent upgrade to Ovi Maps meant that the new maps uploader wanted a phone connected with Nokia PC Suite via the USB-A to Mini-B cable or bluetooth in PCSuite mode.

However without a cable or bluetooth (I have my old IBM X31 laptop, hand upgraded with Intel 2200BG Wifi) what can you do?

Turn to an almost forgotten technology: IrDa. The N95 8GB has a 115200 IrDa port on it and the Thinkpad X31 has a 4Mbps one.

1. Download the old "Nokia Map Loader" - this talks to the phone as a mass storage device.
2. Connect the phone via IrDa to the PCSuite.
3. Run the Nokia Map Loader

Upload the maps - took a very long time!

[An alternative to doing this is use the following blog post to download the maps directly. This can even be done on your mobile!]

Off to Microelectronic Systems Education

...in sunny San Francisco.

MSE Conference (and its partner EWME) provide a fantastic venue to explore and demonstrate novel ways to teach complex system-on-chip designs.

I shall also be attending the tutorials - One on the Cypress PSoC platform, a fascinating combination of a digital microcontroller, a programmable logic device and a section of programmable analogue logic. The second will be on several techniques from Synopsys:

1. Structured Methodology for Successful Low Power Verification
2. Power-Performance Optimization of Digital Circuits

Fix for amanda 2.5.1p1 on MacOSX PowerPC

Bug causes the following error message in the following stable releases 2.5.1p1, 2.5.1p3, 2.6.1p1

Amanda Backup Client Hosts Check
--------------------------------
ERROR: NAK mac-mini: user backup from hs-dhgl7ce.msindustries.com is not
allowed to execute the service noop: /Users/amandabackup/.amandahosts:
owned by id 5000, should be 529288
Client check: 1 host checked in 0.859 seconds, 1 problem found

There is no user of UID of 529288 on the machine.

Here is a snip of the amandad log after a modification to add Print statements printing the UID as dereferenced from the pointer in check_user_amandahosts() in common-src/security-util.c

amandad: time 0.187: sec: calling accept_fn
Print uid: 5000
Print UID on entry now: 5000
check_user_amandahosts(host=hs-dhgl7ce.msindustries.com, pwd=0x100400, remoteuse
r=backup, service=selfcheck)
amandad: bsd: processing file: /Users/amandabackup/.amandahosts
amandad: bsd:                  owner=amandabackup group=wheel mode=600
Print UID after stralloc: 529288
Print UID after fopen: 529288
Print sbuf.st_uid now: 5000
Print UID after fstat: 529288
amandad: bsd: processing line: hs-dhgl7ce.msindustries.com backup="" amdump=""
amandad: bsd: comparing "hs-dhgl7ce.msindustries.com" with

The bug has been fixed by caching the pwd->ps_uid at the start of the function and using the test against that. Also note that the bug behavior changed at some point as the UID changed from 529288 to 0. Suggests a buffer overrun, endianness or thread issues.

Here is the modified check_user_amandahosts() security-util.c (don't forget to modify the top of the file to enable secprintf().

/*
 * Check to see if a user is allowed in.  This version uses .amandahosts
 * Returns -1 on failure, or 0 on success.
 */
char *
check_user_amandahosts(
    const char *        host,
    struct in_addr      addr,
    struct passwd *     pwd,
    const char *        remoteuser,
    const char *        service)
{
    char *line = NULL;
    char *filehost;
    const char *fileuser;
    char *ptmp = NULL;
    char *result = NULL;
    FILE *fp = NULL;
    int found;
    struct stat sbuf;
    char n1[NUM_STR_SIZE];
    char n2[NUM_STR_SIZE];
    int hostmatch;
    int usermatch;
    char *aservice = NULL;

    uid_t tmp_pw_uid = pwd->pw_uid;
    secprintf(("Print UID on entry now: %d\n",pwd->pw_uid));

    secprintf(("check_user_amandahosts(host=%s, pwd=%p, "
                "remoteuser=%s, service=%s)\n",
                host, pwd, remoteuser, service));

    ptmp = stralloc2(pwd->pw_dir, "/.amandahosts");
#if defined(SHOW_SECURITY_DETAIL)                               /* { */
    show_stat_info(ptmp, "");;
#endif                                                          /* } */
    secprintf(("Print UID after stralloc: %d\n",pwd->pw_uid));
    if ((fp = fopen(ptmp, "r")) == NULL) {
        result = vstralloc("cannot open ", ptmp, ": ", strerror(errno), NULL);
        amfree(ptmp);
        return result;
    }
    secprintf(("Print UID after fopen: %d\n",pwd->pw_uid));

    /*
     * Make sure the file is owned by the Amanda user and does not
     * have any group/other access allowed.
     */
    if (fstat(fileno(fp), &sbuf) != 0) {
        result = vstralloc("cannot fstat ", ptmp, ": ", strerror(errno), NULL);
        goto common_exit;
    }
    secprintf(("Print sbuf.st_uid now: %d\n",sbuf.st_uid));
    secprintf(("Print UID after fstat: %d\n",pwd->pw_uid));

    /*if (sbuf.st_uid != pwd->pw_uid) {*/
    if (sbuf.st_uid != tmp_pw_uid) {
        snprintf(n1, SIZEOF(n1), "%ld", (long)sbuf.st_uid);
        snprintf(n2, SIZEOF(n2), "%ld", (long)pwd->pw_uid);
        result = vstralloc(ptmp, ": ",
                           "owned by id ", n1,
                           ", should be ", n2,
                           NULL);
        goto common_exit;
    }