
.DF
                                                         3/28/85
                                                     Rob Horning
.DE

I have spent the last couple of days investigating the risk of depending
on the availability of 25ns static RAM's.  Below is a summary of the
investigation followed by a summary of the data gathered form each vendor.

.H 1 "Summary"

Depending on the availability of 4K x 4 25ns static RAM's for production parts
around 1Q86 has to be considered a low risk.  There are 5 vendors who
expect to have this part in production by 1Q86.  There should be no problem
with getting parts for proto-types because at least one vendor has them
available now (6 week lead time).  I have spec sheets from three of the 5
vendors and all three have the same pin out (20 pin dip).

Because of space, power and cost constraints Firefox would prefer to use
2K x 8 RAM's for the cache and TLB memory.  This part is currently
available (samples) as a 35ns part.  At least one vendor plans on having
a 25ns part sampled by 3Q85.

Hitachi plans on sampling 16K x 4 25ns static RAM's 3Q85 and on being in
production with the part in 1Q86.  We should not depend on getting this
part but we should design the CCU so that it can take advantage of this
part when it is available.

Board space and timing constraints make it desirable to use surface mount
parts.  Most of the vendors plan on doing a surface mount part, but this
part will not be available until after the DIP part.  There is not a standard
surface mount package agreed upon.  We should not depend on getting a surface
mount package.

None of vendors had soft error rate specs or data.  By the time that good data
is available it will be too late to add error correction.  We should assume
that error correction of some kind will be needed.

.H 1 "Further Action"

Assuming that we decide to use commercial static memorys there are several
things that need to be further investigated.  The timing should be designed
assuming 4K x 4 25ns RAM's in a DIP.  The chips should allow for easy expansion
to the 16K x 4 RAM.  We should also make sure we do not do anything that would
prevent us from using 2K x 8 RAM's.

.H 2 "Failure/Error Rates"

We need to investigate further what the failure and error rates are and also
what the failure mechanisms are.  This data is needed to optimize our error
handling strategy.

.H 2 "Other RAM Organizations"

The 16K x 4 and the 2K x 8 static RAM's could improve the Cheetah and Firefox
products.  We should investigate how high of risk these parts are and how much
they help performance and cost.  We should be able to do this without effecting
the CCU and TCU schedules.

.H 2 "Power"

The 16K x 4 RAM's use about .5 to .75 watts/chip.  We need to get better
resolution on what the power will be and how this effects Cheetah and
Firefox.  We need to understand how the other RAM organizations effect
the power.

.H 2 "Package"

The package for the 4K x 4 DIP RAM is not an issue.  The 2K x 8 Dip RAM is
also standard.  We need to understand if the 16K x 4 RAM will have a
standard package.  We should also keep up with the the progress of the
surface mount alternatives.  We should encourage the vendors to converge on
some standard PLCC package.

.H 2 "Japan Visit"

HP sends a team to Japan every year to discuss RAM needs and issues with our
major RAM vendors.  This
trip is planned for mid May.  We need to understand our needs and concerns
and get then documented in the next few weeks so that they can be added to
the agendas for these meetings.  We need to make sure that our concerns will
be well represented.

.H 1 "Vendor Data"

Following is summary of the information gathered from the RAM vendor.
I mention every vendor that was contacted in order to give a clear
picture of where the industry is heading with static RAM's.  Keep in mind
that this data was gathered on short notice and so a lot of it is incomplete
and we need to follow up on the data that we do have.

.H 2 "Cypress"

Cypress is currently sampling 4K x 4 25ns RAM's in fairly large quantities.
They are shipping 800 parts to Logic Systems Division this week.  They say that
they can ship quantities in the thousands in the next couple of months.
They currently have a six week lead time.

.H 3 "Cost"

.DL
.LI
Small quantities - $26.
.LI
10,000/year - $15.
.LI
100,000/year - $10. to $12.
.LE

.H 3 "Package"

They have the standard 20 pin DIP and a 20 pin ceramic chip carrier.  They plan
on having a PLCC later this year, but he was not sure of what package.  The
PLCC will probably cost about 10% more.

.H 3 "Power"

The active power is .5 watts and the standby power is .08 watts.

.H 3 "Error Rate"

They are just getting started on testing the error rate.  The 4K bit part
had no measurable soft error rate.

.H 3 "Other Parts"

They currently have a 2K x 8 35ns RAM.  I did not find out what there plans
are for future products.

.H 3 "Vendor Rating"

Cypress was visited by FSD material engineering last month.  The report
was very favorable.  They are very concerned with quality and have a
good system for tracking it.  We are still waiting for a financial report.

.H 2 "Hitachi"

Hitachi will sample 16K x 4 25ns 3Q85 and be in production 1Q86.  There current
4K x 4 is 35ns and the rep did not know of plans to offer this part in 25ns.
Hitachi is currently sampling a 64K x 1 25ns static RAM also.e 16K x 4
will be available in a PLCC but not until 3Q86.  I have no data on reliability,
package, or power for this part.


.H 2 "Inmos"

They have a 35ns 4K x 4 part that they are starting to bin as 25ns parts.
A 45ns 4K x 4 part is being used by FSD on the series 200U CPU board.  This
part has been in production for over a year.  The 25ns part will be sampled
in about 2 weeks and in production in about 3 months.

.H 3 "Package"

They use the standard 4K by 4 DIP package.

.H 3 "Power"

The 35ns part uses .5 watts active and .1 watts standby.  The 25ns part would
probably be about 20% higher.

.H 3 "Error Rate"

They did not have soft error rate data but there failure rate testing has
shown there hard failure rate to be less than 150 FIT.

.H 3 "Vendor Rating"

They are not considered a good vendor because of delivery and quality problems.

.H 2 "Mitsubishi"

They will have a 4K x 4 20ns sampled in 3Q85 and in production 1Q86.  They are
currently sampling 35ns parts.  The rep still has to get back to me with
data on package, power and cost.  They are one of our best vendors.


.H 2 "Toshiba"

They currently make 35ns 4K x 4 and 2K x 8 RAM's.  They are going to start
binning 25ns parts.  They can start providing samples in about 2 weeks.

.H 3 "Cost"

(These are the rep's estimates.)
.DL
.LI
Small quantities - $10.
.LI
10,000 - $8. to $10.
.LE

.H 3 "Package"

Both parts are offered in the standard DIP packages.  They plan on offering
them in PLCC'c at some time in the future.

.H 3 "Power"

Both part require close to a watt of active power.  The part does not have a
standby mode.

.H 3 "Vendor rating"

Toshiba is considered a good vendor and currently provides many parts to HP.

.H 2 "Motorola"

They now have a 2K x 8 35ns part and expect to be able to sample 25ns parts
in 3Q85.  The 25ns parts will be binned from the 35ns design.

.H 3 "Package"

They have the standard DIP package.

.H 3 "Other Parts"

They are currently sampling an 8K x 8 45ns part.


.H 2 "Intel"

Intel will have a 4K x 4 25ns RAM sampled in 3Q85 and in production 4Q85.
I am not sure what packages will be offered.

.H 3 "Cost"

10,000 - $13.

.H 3 "Power"

The part will require between .5 and .75 watts.  It does not have a standby
mode.

.H 3 "Other Parts"

They will sample a 16K x 1 25ns static RAM in 2Q85.

.H 3 "Vendor Rating"

Intel is not considered a very good vendor by FSD because they do not support
there older products very well.

.H 2 "AMD"

They are just coming out with a 4K x 4 35ns part.  The rep was not sure when
this would be available as a 25ns part.  He will get back to me.  They have
a 1K x 4 25ns tag RAM (24 pin) that he thinks will be available this year in
a 15ns version.  He will get back to me on this also.

.H 2 "Fujitsu"

They have a 4K x 4 35ns part.  The rep does not know of plans to offer a
25ns part.


