Liz and I attended a panel at DesignCon that asked the question: what are you doing about the chip killers that delay your tapeout? That’s an intriguing, possibly unanswerable thought, since we’ve asked that question virtually since EDA’s inception. Ed Sperling of Systems-Level Design moderated the panel which had on it: Sunil Malkani of Broadcom, Ravi Damaraju of Juniper, Ramon Macias of NetLogic, John Busco of NVIDIA and Bernard Murphy of Atrenta.
Sperling moderated a lively discussion; questions that he or the panelists or audience posed highlighted the ongoing nature, or unanswerability of the topic. Some were:
• As designers and design managers, what keeps you up at night?
• If your design has to finish in half the time that your previous project took, do you start with a [design methodology and flow] clean slate?
• How do you get hardware and software engineers to work together?
• What’s good enough to get the design out the door?
• How do you define failure?
• What’s the price of failure?
• Who owns quality?
• What do you do when your next project is 4X the size of your last design? Throw people at it? Make the tools do more? Run faster? How?
• How do I turn around a design in a month and get all of these [now-required] apps on it?
• Why does place & route have to be flat?
• When will P&R, timing analysis have to break down the design hierarchically?
• How can verification be improved so that its pessimistic estimates won’t require designers to over-design?
The panelists all bemoaned the dueling standards that plague EDA, attributing them to companies wanting to gain marketing advantage, to the detriment of EDA users.
Sperling will publish a transcript of this panel in a future issue of System-Level Design. Nic Mokhoff published a summary of the panel the next day.
Finally, I have a question: why does DesignCon schedule a management-level panel on a day when the exhibit floor isn’t open? Doesn’t help DesignCon panels’ attendance, which has been paltry for years, seems to me.
– end –