I design fast (OC48) fibre linecards. The biggest thing that holds us up is part acquisition, and it isn't always in ICs. When it's not RAMs or FIFOs (notably IDT parts), it's passives like low ESR tantalum capacitors.
Our assembly house recently bought a reel of 330 uF tantalum caps for US$15/piece. With an MOQ of 1000, that cost US$15,000, which is more than the average new car in this country! (.nz) That's one reel of capacitors! The guy had about an hour to make the decision to buy before someone else bought it instead.
This problem is not new. It is just getting worse. We have been told tantalums will be impossible to get (for prototyping groups like ourselves) in the next few months.
For some reason, we haven't had difficulty getting flash ROMs like other people have. I put that down to the package type (PLCC) we use.
3) Your SDRAM won't work at 150MHz because it's simply too fast for the junctions to switch, not because of EMI.
Two things limit the maximum clock speed of a digital device: power dissipation and timing constraints.
Typically, one limit will mask another. In the case of PC-class CPUs, it is generally power dissipation that limits the top speed (as most people realise, power dissipation is directly proportional to clock speed) which is why all these weird and wonderful cooling devices make overclocking possible.
Eventually, you'll reach the second limit -- timing constraints. Timing constraints reflect the time taken for electrons to flow along a conductor and charge the capacitive loads they are connected to. They vary only a few percent with temperature, so no practical amount of cooling can overcome this limit.
I have no idea which of these RAMs are limited by. My gut feeling is that they are limited by timing constraints rather than power dissipation (for various reasons, the main one being that RAMs don't really get hot.) Anyone know for certain?
They could probably even cause current routers to start this behavior with a microcode
Firmware patches are useless because modern routers don't have time to process every packet through its CPU. High end routers do most things on the line card, which doesn't involve the router's CPU.
This is why traceroute no longer produces sensible results. ICMP has to be processed using the router's CPU and therefore is given second priority, which is one reason why the RTTs don't always increase through successive hops.
Slashdot Mirror
Our assembly house recently bought a reel of 330 uF tantalum caps for US$15/piece. With an MOQ of 1000, that cost US$15,000, which is more than the average new car in this country! (.nz) That's one reel of capacitors! The guy had about an hour to make the decision to buy before someone else bought it instead.
This problem is not new. It is just getting worse. We have been told tantalums will be impossible to get (for prototyping groups like ourselves) in the next few months.
For some reason, we haven't had difficulty getting flash ROMs like other people have. I put that down to the package type (PLCC) we use.
Two things limit the maximum clock speed of a digital device: power dissipation and timing constraints.
Typically, one limit will mask another. In the case of PC-class CPUs, it is generally power dissipation that limits the top speed (as most people realise, power dissipation is directly proportional to clock speed) which is why all these weird and wonderful cooling devices make overclocking possible.
Eventually, you'll reach the second limit -- timing constraints. Timing constraints reflect the time taken for electrons to flow along a conductor and charge the capacitive loads they are connected to. They vary only a few percent with temperature, so no practical amount of cooling can overcome this limit.
I have no idea which of these RAMs are limited by. My gut feeling is that they are limited by timing constraints rather than power dissipation (for various reasons, the main one being that RAMs don't really get hot.) Anyone know for certain?
Firmware patches are useless because modern routers don't have time to process every packet through its CPU. High end routers do most things on the line card, which doesn't involve the router's CPU.
This is why traceroute no longer produces sensible results. ICMP has to be processed using the router's CPU and therefore is given second priority, which is one reason why the RTTs don't always increase through successive hops.