Jun 30 2009
Cadence Services Formally Addressing the SaaS Model
The service business at Cadence has been steadily operating at over a $100M/yr run rate. Their services group has been in place performing tactical design for clients (chips and IP), tactical CAD support file and technology file/application program file creation and flow/methodology creation. This projects and billings for the group have been organized for milestone based funding releases. The current economy has forced a changed in this model as a rolling cash flow and time based releases have become a big issue for most clients and for the service group’s revenue recognition. This change has driven the Cadence services group to add a SaaS (Software as a Service) model to their offering. The offering and the technology has been developed and deployed on a selective basis for the past 8 years and currently supports a less than 100 clients.
Cadence’s solutions offerings are split into five areas: Systems, Enterprise Verification, Low Power, Mixed Signal and Advanced Nodes. Their SaaS offering is focused on the tactical services offerings that can use defined and stable manufacturing flows and tools – these are Enterprise Verification, Low Power and Mixed Signal. These three areas are being addressed with a service offering that includes an IT configuration (both client and supplier sides), training for both products and design applications, an engineering/task based outsourcing and the design environment/flow. This offering is being brought out as a business unit based on recent survey results showing on a typical design project, 17% of the budget was EDA Tool cost, and 23% was time and resources to develop and implement the production environment and flow PER DESIGN.
The model that is being used is that of a traditional ISV scenario. The service offering is presenting not a pure play SaaS model where the software is just placed in a cloud to achieve higher availability, rather is is an integrated hardware, software and service offering following the model and techniques of ISV leader Oracle. The Cadence solution uses for following mapping of the Oracle SaaS stack model:
* Access and management : Environment management interface and Remote user desktop
* User interface : End-to-end use model(s) [flows, methodologies, best practices, etc]
* Integrated Software : Cadence products and solutions
* Database : Design database storage
* IT Infrastructure : Secure Networking and High performance compute resources
In the new solution the Customer Site only needs to host the design team, a small IT staff and the Cadence remote desktop connected through a VPN. The Cadence side of the VPN is a hub that consists of Managed design and IT services, compute resources, the design database and packaged design environment with software access. These hubs are located in 11 location world wide to insure data transfer and access availablity. These 11 sites are also high availability compute centers with downtimes ranging from 28hrs/year (99.671% avail) to 0.4hrs/year (99.995% avail).
The SaaS programs are project solution based, not tool availability solution based. This is not simply purchase access to X number of simulation runs or a bag of tokens that can be used on more than one tool in a product family. The basis is the leveraging of a standardized design environment (naming conventions, scripting, tool sequence, netlist formats, BIST procedure, etc) for a given foundry/process selection. The use of these standardized environments can lead to schedule improvements of start of design for a full design team in 1 week vs prior project 3 months. This improvement was obtained by a new medical device manufacturer.
The SaaS model is an service level agreement, so levels of support, mutual resource allocation, length of data retention, data recovery and backup and IP access/compatibility with the flow are all areas of contract negotiation for the projects. This offering is available now from Cadence.
PC
This is very interesting. Oracle has met with a lot of success in their hosted model. The standardized flow approach seem very close to the “fabless ASIC model.” It’s more than an experiment if it accounts for more than 10% of Cadence’s revenue–if I am interpreting your report correctly and this contributes more than $100M in revenue.
Couple of questions:
1. Would Amazon’s Elastic Computing Cloud platform represent an option for expansion?
2. How would you re-architect the tools if this became more than 50% of Cadence revenue? Does the Mentor Xtreme PCB model with it’s realtime shared database allow you to “follow the sun” more easily in a global design team? (I am not an employee or affiliated with Mentor).
3. What’s the nature of the “flow management layer” that detects and maintains these high availability levels? Is that something Cadence might also productize to allow internal groups to improve uptime/availability? Does this look more like LSF, Hadoop, or something else?
4. The availability metric seems more appropriate for a telephone network or a transaction processing environment where any one session might last only a few minutes. Absolute outage count might be more useful for assessing impact when you have long jobs. Alternatively, it should be de-rated for average job runtime. For example if you are running a lot of 40 hour P&R jobs one outage on average will cost you 20 hours in lost time.
Thanks for posting this, it was very thought provoking.
Cadence actually announced this offering last September if I am not mistaken. I wrote about it some on my blog back then http://tinyurl.com/ltgcww. I’d be curious if there was something new that they did not offer before.
Sean asks some great questions as well that I might be able to address:
- the $100M is the entire services business, not just SaaS
- As of the last time I spoke to Cadence, they used their VCAD infrastructure, not any cloud computing provider like Amazon. Adding a cloud capability would be fantastic as it would enable much quicker provisioning
This is a follow-up on the September announcement. I reported it to show that they decided to stick with the model after the restructuring at Cadence. The key issues that prompted my discussion on the subject were:
(A) The scope of the service business in total is $100M, the belief at CDN is that this model will help grow that dramatically (with dramatically not being a quantified value).
(B) the SaaS model is predicated on the creation and use of a “standard design flow” and has a mixed signal customer as the reference account. The fact that the largest installed base of custom IC tools and IP creation tools, is now presenting the impression that there is a “usable & standard” process specific design flow for mixed signal is new. The “buzz” at last year’s DAC, which never really materialized in revitalizing the industry, was that a shift towards automated custom design was immanent.
(C) The compute environment that they are promoting is not a cloud based environment, it is basically their VCAD structure from their internal design support groups with a secure external VPN hook. This is like a first stage virtualization that has been taking place in the financial and IT services business. For this reason, and trying to mimic the Oracle biz model, they have been reporting the availability stats on the setup. HA systems in environments with a local power grid and a reasonable 24/7 bandwidth availability are not really new in the IC space, most IDMs require them. It is new to have them available for a small firm to use, as those systems usually cost the same as Series A &B funding combined for a new chip startup. The discussion did include the location of these centers and their strategic placement in order to allow for the interactive tasks to take place without any key latency issues on view. The one point that was not adequately addressed was how the large data sizes (this is not a thin client application model) would work for interactive review of SOC level designs and other statistical / multi-corner simulation / verification results. To date, anyone offering an SaaS model has not been successful with offerings requiring real time manipulation and review of GB level live data. I think – guess on my part – that since their first reference customer is mixed signal, the environment has not been pushed by a GB live data client as yet, but has held up satisfactorily for a MB class mixed signal customer.
(D) The availability profile and IT scenario is being prestented based on traditional metrics of “short query activity vs operating time” of the setup. In actual design activities, there is a mix of long jobs and short jobs, the issue is not the down time, but the state and availability of the metadata and restart mechanism after the downtime. That discussion was not covered in the interview due to lack of time.
Hope this helps clarify things. They fact that they are trying this in the current economy, after the September world financial situation changed is one of the main points of repeating and clarifing the announcement from Sept.
PC
“It is new to have them available for a small firm to use, as those systems usually cost the same as Series A &B funding combined for a new chip startup.”
I think this is a key point. The larger full-flow EDA companies have primarily focused on their top-tier customers. Meanwhile, smaller customers and startups struggle to afford to purchase the tools they need. A SaaS model where the entire flow (CPUs, tools, flow) can be “rented” is very attractive to these smaller companies who can now turn a large up-front capital expenditure into an operating expense. I’d be interested to hear of successes outside of the analog / mixed-signal space.