[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[f-cpu] ERIN32



    I see some points made on circuit loading.  While at Sanders Associates Inc several years back; it was determined that a Logic design having a maximum fan-out of 7 loads had a good chance of surviving temperature testing over the military temperature range of -55 to +125c.  In my design of the Erin32 I took special care to have a maximum load of 8.  In the chip design process; there is an Auto-buffer option - I always use it.
 
    The End-item Processor is a Quicklogic Corp QL6600-7 672 pin Plastic Ball Grid Array device of the Eclipse Family.  The current useage of the device is as follows:
 
            Utilized cells (no buffers)        3048 of    4032
            Utilized cells (buffered)           3537 of    4032
            Clock only Cells                         1 of         9
            Bi Directional cells                  502 of      506
            PLL cells                                   1 of         4X
            Flip-flop of IO cells                  224 of      508
            1st Flip-flop of Logic cells        981 of    4032
            2nd Flip-flop of Logic cells       982 of    4032
            Routing resources               83167 of    293995
            ViaLink resources               73301 of   7957982
            Global Clock Nets                      9 of         9
            OSC1 Clock                        93 MHz
 
    This design is of the M2M variety that you people are steering clear of.  A modified Harvard Architecture is used -
that is separate Program and Data memory also having an 18 bit Source and Destination address, and ONE ACCUMULATOR REGISTER.  Eight (8) Instructions and eight (8) operands are fetched and executed in parallel if Data Waits do not interfer.  This function was also shot down by you people - since I have used this feature some time ago (before most of you were born); I used it again, instead of using Cache.  To me it appears that 8 instructions is a good number because one of eight is probably of the Jump or Conditional variety anyway.  This is borne out of the analysis I did a couple years back - and the numbers are approximately correct.
    The design is all Schematic Level, so I have complete visability of every GATE. Although the Two and Four Port SSRAM's have an access time of 4.5 NS; I round this to 5, and 5 NS setup, and 5 NS Ram to CPU I/O pin.
    Multiply and Divide are included; however Square Root had to be abanded due to an insufficient number of Logic cells. This will require a separate optional Chip. This will fit very well in the Optional Floating Point Processor.
 
    I will be very happy to answer any questions of the design.
 
Regards
Richard E. Hartney
Research Director
Erin Greene & Associates
E-mail  rhartney@bellsouth.net