We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.
I am a junior ASIC engineer. At the moment, I have three the question about AMBA AXI 3.0
1. No combination paths between master and slave, all timing path must be registered. Why the VALID and READY signals can assert in the same time ?
2. In the FIXED burst address, If Start_Address = 0x48, Burst_length = 4, Burst_size = 4 --> can you calculate for me ?
3. If I have a FIXED bursts address, Can I read FIFO 8 times in a row ?
4. What the different between READ ADDRESS channel and WRITE ADDRESS channel ? can it combine into one channel ?
If you have figure illustrate, please show me !!!
Thank you so much.
Hello,
Between a master and a slave (i.e. VALID and READY) should not be always registered. In the case VALID is output and READY is input from a master side, when READY is activate within one CLK cycle after VALID was activated, it is the situation that the VALID and READY assert at the same tame. This case may happen READY is made directly from VALID.
It is the same as such a transaction of which address is 0x48 and of which size is 4 bytes is repeated 4 times.
It is true if the burst length is 8.
Regarding 'in a row', I cannot understand the meaning.
AXI read channel and write channel are independent (should be work concurrently) and it is impossible to combine signals of the read and write channels.
However, Read-After-Write (i.e. the data dependency of the same address between the read and write channel) should be considered.
Best regards,
Yasuhiko Koumoto.
thank for your help yasuhikokoumoto
Hi yahushikokoumoto,
Between a master and a slave (i.e. VALID and READY) should not be always registered
In the chapter 3.1 of the specification is "There must be no combination paths between input and output signals on both master and slave interfaces"
the statements only say that a master and a slave must be able to work independently.
For example, READY input cannot be registered.
If it was registered, the de-assertion of VALID would delay and a salve misunderstood it as that one more request came.
Unless the VALID could be directly forwarded to the READY logic, the Figure 3-3 timing (it is also the same as shown by me) might impossible.
Thank Koumoto so much, I have already understood .