From 772@??? Sat Jan 10 21:27:57 1998
Received: from pe-gtate rambus.com (pe-staterambus.com (192.93.42.39))
‘by jupiter rambus.com (8.8.5/hub-sunos) with SMTP id MAA2S197:
Sat, 10 Jan 1998 12:18:39 -0800 (PST)
Message-Id: <3.0 32,199801 10121411 006¢5S80@198.147 244.4
XeSender:gtateq@198.147 245.4
X-Mailer’ Windows Eudora Pro Version 3.0 (32)
Date: Sat, [0 Jan 1998 12:14:17 -0800
To: exee@rambus.com
From: Geoff Tate
‘Subject: strategy re belot
Ce grate@rambus com, Neming@rambus.com, champel@rambus com,
abealocerambus.com, chen(@rambus com. naono(@rambus com,
Groberis@rambus com, barth@rambus.com, dmuthani@rambus.com
Mime-Version: 1.0
Content-Type: texv plain; charset="as-asci”
N-UIDL, $34a9b76a7e454007
here's some brainstorm ideas on strategy for minimizing the negative impact
of blot. the following is “NOT* official strategy, just my saturday
‘morning thoughts:
die size story to intel
got to get back to intel by end january latest, as we've commuted, with
5) specific recommendation fr low cos altemative architecture for camino
to support
>) aie size story, that convinces pete, that for curent archilower cost
lternative we can be <10% and hopefully <5% versus SDRAM even forthe
shrink paths
then we need to get this same story infront of all the dram engineering
managersiexecs fo convince them.
fd take a simplified version to customers‘analyst.
engineering support allocation
‘We have to meet our contractual commitments; and we don't want to pis off
‘dram partners - but within these constrains i agree with dave r that we
‘Should stan to prioritize who we provide support (above and beyond what
the contract calls fr) to and priontize based on who is aggressively
‘promoting rambus and quoting aggressive prices for rambus
‘get some early deals done for aggressive parmers?
{ave said jay belUdel is willing to place a “major ordet” for direct
ambus. inel i thinking f LG is walling t ake direct rdram orders at
the same price as sdram then werintl should try to work out a marriage
between Dell and LG to get such an order placed. this would presumably be
big win for LG: business and strategic relationships, which is why they
fre promoting rambus - to get an edge on samsung/etc. and it would be 2
‘wakeup call for samsung/ete about the risks of taking the high price road.
RH 386371
R 233877
OUTSIDE COUNSEL ONLY
‘cx0987-001exec callsipressure
i probably need to do another round of meetings with exees of top companies
ta get them to review their direct rambus pricing story and put mace
pressure on them (but not so much as to demotivate them to be more
aggressive, especially NEC where we have good strategic relationships.
also we need to review their specific plans for investment for
‘esvproduction as these plans should be geting fixed later this quarter
for the next 12 moths
controLroyalties on beloit
\we should get patents/contracslete in place so that dram companies/module
companies'pe companies cannot builduse beloit without legal agreements
with us. then we have leverage to get royalties on beloit in return,
Redacted Information
also the dram companies who do not havea logic license with us will need
‘one if they want to build their own beloit (which i think is highly likely
for them to minimize costsluse their own fabs which are empry). in these
negotiations we can get
4) patent pooling fom samsung
') commitments to do a non-belit logic product of sign
‘efforts basis as a condition
¢) higher royalty rate for belot application - say $-10% (33.5% for other
rambus logic)
4) minimum zoyalty per chip - say $0 cents or 25 cents
¢) minimam royalty per quarter
£) also since beloit is only going to show up on the module there won't be
1 "market selling price” - we need to address that somehow so we dont get
screwed with artificially low transfer prices.
ce on best
Redacted Information
1 think we should get betoit called something like Rambus SDRAM Module or
RIMML-S or RIMM-133 with arambus logo on it and have consumers hear about
the RAMBUS channel with RAMBUS memory modules tht include RIMM-133,
RIMM-600, RIMM.800.
beloit modules will need as much support and handholding for
chipimodule/system companies as RIMMs, sho will provide the support? my
RH 386372
R 233878
‘OUTSIDE COUNSEL ONLY
‘cx0987-002guess i intel would like us to doit. 4 think we should and take eloit
{nas “one of our own” and provide the same level of support as for RIMM.
bur we need to get compensated fori. the deal with intel was no
incremental royalties on their chips but we'd make SS on rdrams, with
beloit we dont get to make money on rdrams but we need to provide suppor.
50 we toll intel we will use the above levers asa way to make say
‘SU/modile in tum forthe support.
(nel says belot will cost 5 to build, we get paid on what i sells,
for. at $5 and average 3.5% royalty weld get 17 cents per beloit module:
at $10 we'd get 35 cents per beloit module. ifthe enure PC market in
2001 is 150M PC's and each uses 1.5 modules, then weld make roughly
S37Miyear ifbeloit is $5 asp and $7SMiyear iPbeloit is $10 asp).
postbeloit-diselosue spin contol to chip parmerssystem.
co.sianalysts investors
wwe need a presentation that examines the impact of beloit and paints a
realistic and positive picrre. (ll ry to take a cut at this pitch this
weekend)
cover points like:
4) ditectrambus dram cost is not going tobe 40% in mass production.
review the current cost story on di sizepkg/module. point out initial
emphasis is cost to market So parts are not optimized. initial die size
5-17% large than sdram. initial total cost maybe 10-25% larger than sam.
‘depending on the manufacturer. story on how we will work ith
‘manufacturers on shrinks and architecture mods to get die size <5% versus
sdram. story on how cost of csp will come down with éarning curve/volume
story/schedule for direct rambus dram cost being <10% in 2000. point out
hhow 11+ suppliers will ensure price tracks cost and there cannot be value
pricing aftr the initial demand ramp phase.
+) cost of beloit: 133mhz sdrams (may) cost more than 100mba sdrams as they
are not the sme mask set and have to source alittle more bandwidth;
tighter Uo specs at 2.5v vdd; same module cost as RIMMs; beloit chip $5-10
adder. so our competition i no longer sdram dimim but beloit and bel
costs $-10% more
«performance of beloit vs rimm-600/rimm-800: latency and effective
bandwidth, beloit is 1/3 the performance. ifbeloit is “good enough” then
‘why did intel bother going to rambus inthe first place. ifbeloit is good
‘enough that implies that intl’ processors either arent going to continue
up a 2x every 18 months curve andlor they be performance starved.
‘either way this isa terrible message for intel. hence intel has to have
continually more bandwidth, but direct rambus provides 3x over sdram-100
and it isnt really needed for most apps for katmai/eamino; but as apps use
up the performance and processors seale up and agp takes off the bandwidth
will be needed.
4) show that beoit cannot scale upto provide rimm-800 performance. show
performanceicost of beloit-ddr. dar sdrams (per intl) are 5% more
‘expensive than sdrams (presumably due to larger cores for 2x bandwidth),
RH 386373
R 233879
‘OUTSIDE COUNSEL ONLY
¢x0987-003